Why Google Dominates Advertising Markets Competition Policy

Why Google Dominates Advertising Markets

Competition Policy Should Lean on the

Principles of Financial Market Regulation

Dina Srinivasan

* Since leaving the industry, and authoring The Antitrust Case Against Face-

book, I continue to research and write about the high-tech industry and competition,

now as a fellow with Yale University’s antitrust initiative, the Thurman Arnold Pro-

ject. Separately, I have advised and consulted on antitrust matters, including for news

publishers whose interests are in conflict with Google’s. This Article is not squarely

about antitrust, though it is about Google’s conduct in advertising markets, and the

idea for writing a piece like this first germinated in 2014. At that time, Wall Street was

up in arms about a book called FLASH BOYS by Wall Street chronicler Michael Lewis

about speed, data, and alleged manipulation in financial markets. The controversy

put high speed trading in the news, giving many of us in advertising pause to appre-

ciate the parallels between our market and trading in financial markets. Since then, I

have noted how problems related to speed and data can distort competition in other

electronic trading markets, how lawmakers have monitored these markets for con-

duct they frown upon in equities trading, but how advertising has largely remained

off the same radar. This Article elaborates on these observations and curiosities. I am

indebted to and thank the many journalists that painstakingly reported on industry

conduct, the researchers and scholars whose work I cite, Fiona Scott Morton and Aus-

tin Frerick at the Thurman Arnold Project for academic support, as well as Tom Fer-

guson and the Institute for New Economic Thinking for helping to fund the research

this project entailed. I remain deeply grateful to the many scholars that generously

shared their feedback, comments, and insight, often on matters well outside of my

area of expertise: Eric Budish, Kevin Haeberle, Scott Hemphill, Michael Kearns, Lina

Khan, Jonathan Macey, Doug Melamed, John Morley, Gabriel Rauterberg, Thomas

Philippon, Marc Rotenberg, Ashkan Soltani, and Chester Spatt, amongst others. I also

warmly thank John Schwall and Rick Arney for helping me to better appreciate the

parallels to financial markets, Zach Edwards for critical thinking and research assis-

tance, Jennifer LaCosse for careful edits and thoughtful suggestions, Chaaru Deb for

excellent legal research assistance, and the editors at the STANFORD TECHNOLOGY LAW

REVIEW for outstanding editorial support. The views and errors therein are my own.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

24 STAN. TECH. L. REV. 55 (2020)

ABSTRACT

Approximately 86% of online display advertising space in the U.S. is bought

and sold in real-time on electronic trading venues, which the industry calls “ad-

vertising exchanges.” With intermediaries that route buy and sell orders, the

structure of the ad market is similar to the structure of electronically traded fi-

nancial markets. In advertising, a single company, Alphabet (“Google”), simul-

taneously operates the leading trading venue, as well as the leading intermedi-

aries that buyers and sellers go through to trade. At the same time, Google itself

is one of the largest sellers of ad space globally. This Article explains how Google

dominates advertising markets by engaging in conduct that lawmakers prohibit

in other electronic trading markets: Google’s exchange shares superior trading

information and speed with the Google-owned intermediaries, Google steers buy

and sell orders to its exchange and websites (Search & YouTube), and Google

abuses its access to inside information. In the market for electronically traded

equities, we require exchanges to provide traders with fair access to data and

speed, we identify and manage intermediary conflicts of interest, and we require

trading disclosures to help police the market. Because ads now trade on electronic

trading venues too, should we borrow these three competition principles to pro-

tect the integrity of advertising?

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

TABLE OF CONTENTS

I. INTRODUCTION ........................................................................................... 58

II. ELECTRONIC TRADING MARKETS .............................................................. 69

A. Advertising Market Reflects the Structure of an Electronic Trading

Market ..................................................................................................... 69

B. Common Competition Problems in Electronic Trading Markets ........... 77

C. The Competition Principles We Apply in the Equities Trading Market 80

III. GOOGLE DOMINATES ONLINE ADVERTISING MARKETS BY ENGAGING IN

CONDUCT LAWMAKERS PROHIBIT IN OTHER ELECTRONIC TRADING

MARKETS ..................................................................................................... 86

A. Google Has Information and Speed Advantages .................................... 88

1. Google Acquires Leading Ad Server DoubleClick ........................... 88

2. DoubleClick Ad Server Starts to Play Favorites When Sharing User

Identity ............................................................................................. 94

3. Google-Owned Intermediaries Have an Information Advantage on

Google’s Exchange ........................................................................... 98

4. Consumer Privacy Offered as Reason for Information Asymmetry

....................................................................................................... 102

5. Google-Owned Intermediaries Have a Speed Advantage on Google’s

Exchange ........................................................................................ 107

B. Discriminatory Routing of Orders and More Speed Races .................. 117

1. Routing Orders to Google’s Exchange and Owned Properties ..... 117

2. Market Creates Invention to Circumvent Routing Restrictions and

Set Own Speeds ............................................................................. 127

3. Google AMP Speed Protocol Restricts Trading Through Non-

Google Venues ................................................................................ 132

4. Google Search “Speed Update” Further Restricts Trading Through

Non-Google Venues ....................................................................... 144

C. Inside Information Abuses .................................................................... 149

1. Google’s Ad Server Shares Information About Competitors’ Trading

Activity with Google’s Exchange and Buying Tools, Permitting

Them to Trade Ahead of Orders .................................................... 149

2. Google Amends Terms and Conditions to Breach “Ethical Walls”

....................................................................................................... 154

IV. POLICY CONSIDERATIONS ........................................................................ 158

A. Advertising Exchanges Should Provide Fair Access to Information and

Speed ..................................................................................................... 158

B. Steps Toward Identifying and Managing Intermediary Conflicts of

Interest .................................................................................................. 162

1. Structural Separations ................................................................... 162

2. Conduct and Disclosure Rules ...................................................... 163

3. Transparency and Disclosure ........................................................ 171

V. CONCLUSION ............................................................................................. 172

VI. APPENDIX .................................................................................................. 174

A. Timeline ................................................................................................ 174

B. Screenshot of Auction Timestamp Transparency ................................. 175

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

I. INTRODUCTION

The business of advertising has changed drastically over the last two

decades. In the past, advertising contracts were negotiated in person—

think Mad Men-type advertising and publishing executives over two-

martini lunches off Madison Avenue. Today, the largest category of ad-

vertising, online advertising, is rarely negotiated by people at all. Ad-

vances in technology allow ad space to be bought and sold electronically

through centralized trading venues at high speeds, without people ever

meeting face-to-face. When a user visits a website, the ad space on a page

is instantly routed into one or more of these venues. There, the space is

auctioned in real-time to the highest bidder. At the conclusion of these

auctions, the advertisers’ ads return and display to the user in time for

the page to load and before the user has noticed anything has occurred.

The user just sees ads targeted to them, say one for Barclays bank.

The rise of electronic ad trading, widely known today as “program-

matic advertising,” paralleled the rise of electronic trading across various

sectors of the economy. In 2005, the New York Stock Exchange merged

with an electronic trading company, sunsetting the buying and selling of

stock on its iconic trading floor on Wall Street.

Around this time, early

advertising technology company Right Media launched the RMX “adver-

tising exchange,” the first-ever electronic trading venue for ads.

Just like

that, by “borrowing tactics from Wall Street,” advertising went from be-

ing a relationship business to a commodity business, with publishers and

advertisers transacting with each other in an electronic spot market.

. NYSE Approves Merger with Electronic Trading Company, N.Y. TIMES (Dec.

7, 2005), https://perma.cc/QR3Y-KMX2.

. Adrianne Jeffries, How to Succeed in Advertising (and Transform the Inter-

net While You’re at It), N.Y. MAG.: INTELLIGENCER (May 31, 2018),

https://perma.cc/W7CG-4LLA (discussing the history of Right Media).

. Stephanie Clifford, Leftover Ad Space? Exchanges Handle the Remnants,

N.Y. TIMES (July 28, 2008), https://perma.cc/UDC4-Q7LN.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

The efficiencies promised by this new way of trading caught on like

wildfire. By late 2009, the RMX exchange, which Yahoo! acquired in 2007

for $680 million, was processing 9 billion ad spaces daily.

Since then, the

percentage of ads traded in this fashion has steadily increased. In 2021, it

is estimated that 87.5% of all online display ad space in the United

States—including that belonging to news publishers such as The Wash-

ington Post and The Des Moines Register—will trade programmatically.

Since the advent of electronic ad trading, however, the market has

become less competitive. At first, the biggest names in tech—including

Microsoft, Yahoo!, AOL and Alphabet (“Google”)—competed vigorously

with each other. These tech companies initially provided sellers (e.g.,

publishers like newspapers) and buyers (i.e., advertisers) with more

choices when deciding which exchanges and other trading middlemen to

use.

Today, a single company, Google, simultaneously operates the lead-

ing exchange and the leading middlemen (i.e., intermediaries) that pub-

lishers and advertisers must use to trade.

. Yishay Mansour et al., Doubleclick Ad Exchange Auction 2, (2012) (un-

published manuscript), https://perma.cc/T5L4-JS4Y (noting Right Media average

daily trades).

. Lauren Fisher, US Programmatic Ad Spending Forecast 2019, EMARKETER

(Apr. 25, 2019), https://perma.cc/852B-PRUA (estimating 87.5% of online display

ads will trade programmatically in 2021). Note, websites can sell their ad space

in open or private exchanges. Data from news publisher trade association Digital

Content Next suggests that approximately 75% of the inventory of large and mid-

dle-tier U.S. publishers currently trades in open exchanges. One would imagine

that the percentage increases if one includes small and local publishers across the

heartland. See Jason Kint (@jason_kint), TWITTER (May 12, 2020),

https://perma.cc/2NPY-SEZT.

. For competition in the exchange market in the late 2000s, see

AdExchanger Staff, infra note 33 and further discussion in Part III.

. The company formerly known as Google was renamed Alphabet Inc. in

2015, at which point Google became a subdivision of the parent organization.

Indirect proof of market power in the form of market share information is noto-

riously difficult to construct with publicly available information. However, sev-

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

In addition to the market becoming more concentrated, it exhibits

characteristics that would trigger concerns in other electronic trading

markets: market growth is distorted, trading costs—between 30% to 50%

of the trade—are high and non-transparent, and conflicts of interests

abound.

For example, on top of operating the largest exchange, as well

eral public reports on market shares support the proposition that Google oper-

ates the leading exchange, buy-side software, and sell-side software in the mar-

ket today. See COMPETITION AND MKTS. AUTH., ONLINE PLATFORMS AND DIGITAL

ADVERTISING: MARKET STUDY FINAL REPORT 20 (2019), https://perma.cc/3GKR-

YJ5J [hereinafter CMA FINAL REPORT] (estimating that in the UK, Google has a

50-60% share of the advertising exchange market, 90+% share of the ad server

sell-side software market, and 50-60% of the enterprise DSP buy-side software

market); Data Processing in the Online Advertising Sector, Opinion No. 18-A-03,

AUTORITÉ DE LA CONCURRENCE, at 86 ¶ 218 (Mar. 6, 2018), https://perma.cc/X7ZQ-

NM6U [hereinafter AUTORITÉ DE LA CONCURRENCE] (finding that Google’s enter-

prise DSP buy-side software—previously called DoubleClick Bidding Manager

or DBM but since renamed DV360—generates the most revenue and has signifi-

cant growth and that Google “has held a leading position” in the ad serving sell-

side software market since its acquisition of DoubleClick); Joe Mandese, Google

Discloses Results of “Exchange Bidding,” Boosts Publisher Yield >40%, MEDIAPOST:

DIGIT. NEWS DAILY (Feb. 16, 2018), https://perma.cc/GN5X-N8ZZ (stating that the

DoubleClick ad server is “by far the dominant ad server used by advertisers,

agencies and digital publishers”); Keach Hagey & Vivien Ngo, How Google Edged

Out Rivals and Built the World’s Dominant Ad Machine: A Visual Guide, WALL ST. J.

(Nov. 7, 2019), https://perma.cc/DKC7-D5MA (reporting that Google’s exchange,

selling tools, and buying tools are the leading ones in the market and stating that

“[m]ore than 90% of large publishers use the Google ad server, DoubleClick for

Publishers, according to interviews with dozens of publishing and ad execu-

tives”); Allen Grunes, Google’s Quiet Dominance Over the “Ad Tech” Industry,

FORBES (Feb. 26, 2015), https://perma.cc/P84T-7GX3 (discussing the leading posi-

tion of Google’s buying tools Google Ads and DoubleClick Bid Manager, which

is now called DV360).

. Several sources deal with the question of trading costs. See Alex Barker,

Half of Online Ad Spending Goes to Industry Middlemen, FIN. TIMES (May 5, 2020),

https://perma.cc/Z3H4-JLBM; Ross Benes, Why Tech Firms Obtain Most of the

Money in Programmatic Ad Buys, EMARKETER (Apr. 16, 2018),

https://perma.cc/96RE-5HMA (industry analysts from Warc estimating interme-

diaries collectively charge 55% of programmatic spend worldwide, based on data

shared by advertising agency Magna Global).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

as the largest intermediaries trading on its exchange, Google has another

conflict of interest.

Google not only sells ad space belonging to third-

party websites, it sells ad space appearing on its own sites, Google Search

and YouTube. When a small business uses Google’s intermediary tool,

called Google Ads, to bid on and purchase ad space trading on ex-

changes, this tool steers that advertiser towards which ad space to buy.

The effects of this conflict of interest are predictable. In 2007, approx-

imately 64% of Google advertising revenue went to Google properties,

including Google Search and YouTube. The remaining portion went to

non-Google properties, like The Post and The Register, that also sell their

ad space through Google’s intermediary tools and exchange.

Almost

. Note, according to some estimates, Google’s buy-side intermediaries—

the buying tools that small and large advertisers use to trade—account for the

plurality if not the majority of buying volume on Google’s exchange. See Kean

Graham, How to Increase Auction Pressure in Ad Exchange, MONETIZEMORE (July 8,

2016), https://perma.cc/BA8B-2YZV (stating that Google Ads, formerly known as

AdWords, is “currently the largest buyer of inventory on the [Google] Ad Ex-

change”); Hagey & Ngo, supra note 7 (stating that media company News Corp.

did not switch from Google to a rival intermediary, because doing so would jeop-

ardize 40% to 60% of the demand the publisher receives in Google’s exchange

from Google’s proprietary demand, Google Ads).

. In Google’s annual 10-K SEC filings, Google breaks down its advertis-

ing revenue as going to “Google properties” or “web sites of Google Network

members.” The term “Google Network members” refers to non-Google websites

on which Google places advertising. In its 2017 10-K, Google explains that it gen-

erally accounts for third-party revenue on a gross basis: “For ads placed on

Google Network Members’ properties, we evaluate whether we are the principal

(i.e., report revenues on a gross basis) or agent (i.e., report revenues on a net ba-

sis). Generally, we report advertising revenues for ads placed on Google Net-

work Members’ properties on a gross basis, that is, the amounts billed to our

customers are recorded as revenues, and amounts paid to Google Network Mem-

bers are recorded as cost of revenues. Where we are the principal, we control the

advertising inventory before it is transferred to our customers. Our control is ev-

idenced by our sole ability to monetize the advertising inventory before it is

transferred to our customers, and is further supported by us being primarily re-

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

every year since 2004 this split has widened, in Google’s favor.

By Q1

2020, the share going to Google properties had increased to 85%.

The

sponsible to our customers and having a level of discretion in establishing pric-

ing.” In 2004, Google buying tools allocated approximately 50% of advertising

revenue to Google's proprietary properties, such as Search, and the other 50% to

non-Google websites selling their ads through Google's buying tools and adver-

tising exchange. Google Inc., Annual Report (Form 10-K) (Mar. 30, 2005),

https://perma.cc/5A4Y-8EY4. It was in 2006 that Google acquired YouTube. An-

drew Ross Sorkin & Jeremy W. Peters, Google to Acquire YouTube for $1.65 Billion,

N.Y. TIMES (Oct. 9, 2006), https://perma.cc/5TG8-8BVE. In 2005, Google's share of

advertising revenue increased to, approximately, 55%; 2006, 60%; 2007, 65%;

2008, 68%; 2009, 68%; 2010, 68%; 2011, 71%; 2012, 71%; 2013, 73%; 2014, 75%; 2015,

77%; 2016, 80%; 2017, 81%, 2018, 82%; 2019, 84%. Google Inc., Annual Report

(Form 10-K) (Mar. 16, 2006), https://perma.cc/Y272-BRAP; Google Inc., Annual

Report (Form 10-K) (Mar. 1, 2007), https://perma.cc/H4ZJ-FL7B; Google Inc., An-

nual Report (Form 10-K) (Feb. 15, 2008), https://perma.cc/W6FU-AA2T; Google

Inc., Annual Report (Form 10-K) (Feb. 13, 2009), https://perma.cc/5PZY-UZS5;

Google Inc., Annual Report (Form 10-K) (Feb. 12, 2010), https://perma.cc/7B6E-

REEV; Google Inc., Annual Report (Form 10-K) (Feb. 12, 2011),

https://perma.cc/9ZKX-XPKL; Google Inc., Annual Report (Form 10-K) (Apr. 23,

2012), https://perma.cc/YS3R-TLE4; Google Inc., Annual Report (Form 10-K) (Jan.

29, 2013), https://perma.cc/3W45-M9R9; Google Inc., Annual Report (Form 10-K)

(Feb. 11, 2014), https://perma.cc/79A2-6TCT; Google Inc., Annual Report (Form

10-K) (Feb. 6, 2015), https://perma.cc/7DJZ-FD8S; Google Inc., Annual Report

(Form 10-K) (Feb. 11, 2016) https://perma.cc/EU2M-T6QC; Alphabet Inc., Annual

Report (Form 10-K) (Feb. 2, 2017), https://perma.cc/4QKP-UUZJ; Alphabet Inc.,

Annual Report (Form 10-K) (Feb. 5, 2018), https://perma.cc/22HL-SSSP; Alphabet

Inc., Annual Report (Form 10-K) (Feb. 4, 2019), https://perma.cc/ELZ2-AC93; Al-

phabet Inc., Annual Report (Form 10-K) (Feb. 3, 2020), https://perma.cc/RWE8-

27PB. For a visual graph of this split, see Appendix A.

. Google Inc. (2005-2016), supra note 10; Alphabet Inc. (2017-2020), supra

note 10.

. See Alphabet Inc., Quarterly Report (Form 10-Q) (March 31, 2020),

https://perma.cc/7872-G8ZW. In this filing, Google breaks down advertising rev-

enue as revenue for “Google properties” and “Google Network Member Proper-

ties.” Here, Google properties includes “Google Search & other properties and

YouTube.” Id. at 33. Google goes on to say that, “Google Search & other consists

of revenues generated on Google search properties (including revenues from

traffic generated by search distribution partners who use Google.com as their

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

lion’s share of Google’s $134 billion in advertising revenue went to

Google’s own.

Problems of distorted growth then extend across the market. Over

the last ten years, the online advertising market has enjoyed double-digit

year-over-year growth.

However, the majority of advertising revenue

and growth has gone to large firms like Google and Facebook that both

sell their own ad space and simultaneously run an electronic market-

place.

Transparency in the advertising market is also minimal. When small

businesses use the Google Ads tool to bid on ad space belonging to third-

party publishers from Google’s exchange, Google does not disclose to

them the price that the ad space actually cleared for and it appears Google

default search in browsers, toolbars, etc.) and other Google owned and operated

properties like Gmail, Google Maps, and Google Play; YouTube ads consists of

revenues generated primarily on YouTube properties; and Google Network

Members’ properties consist of revenues generated primarily on Google Net-

work Members’ properties participating in AdMob, AdSense, and Google Ad

Manager.” Id.

. PricewaterhouseCoopers, Internet Advertising Revenue Report: 2019 First

Six Months Results, INTERACTIVE ADVERT. BUR. (Oct. 2019), https://perma.cc/YX8X-

72EC (displaying year-over-year growth by quarter from 1996 through Q2 2019).

. Lauren Fisher, Digital Display Advertising 2019, EMARKETER (Jan. 22,

2019), https://perma.cc/4UVF-JGHX (estimating that Google and Facebook will

account for 52% of the display online advertising market in the U.S. in 2019);

Michael Barthel, 5 Key Takeaways About the State of the News Media in 2018, PEW

RSCH. CTR. (July 23, 2019), https://perma.cc/SJ25-45VC (summarizing that digital

ad revenue has grown exponentially but that the majority goes to Google and

Facebook). Outside of Google, Facebook, and Amazon, some even estimate that

the online advertising market is shrinking. See, e.g., Peter Kafka, Google and Face-

book Are Booming. Is the Rest of the Digital Ad Business Sinking?, VOX (Nov. 2, 2016),

https://perma.cc/QAZ6-6V9K (reflecting comments by industry analyst Brian

Weiser and publisher trade association executive Jason Kint). Finally, for docu-

mentation on the auction marketplaces run by Facebook and Amazon, see Face-

book Audience Network, Solutions Overview, https://perma.cc/KYK5-HPUG (last

visited Sept. 29, 2020). See also Amazon Publisher Services, Unified Ad Marketplace,

https://perma.cc/82YF-AX7B (last visited Sept. 29, 2020).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

can arbitrage advertisers’ bids across two Google-controlled market-

places—a fact that may go unnoticed by these small mom-and-pop busi-

nesses due to the complexity of Google’s terms.

In effect, the counter-

party to these advertisers is often Google, though they may be under the

illusion that Google is their agent. At the same time, Google does not dis-

close to the publishers on the other ends of these trades what their space

ultimately sold for and how much Google keeps as its share.

. Google explains how its auctions and pricing work across multiple,

different documents. See Google Ads Help, How the Google Ads Auction Works,

GOOGLE, https://perma.cc/6NV9-43XK (last visited Sept. 29, 2020); Google Ads

Help, How Google Ad Manager Works with Google Ads, GOOGLE,

https://perma.cc/X6UQ-AN7S (last visited Sept. 29, 2020); Google Ads Help,

About the Display Network Ad Auction, GOOGLE, https://perma.cc/8P3M-VPJ3 (last

visited Sept. 29, 2020); Google Ads Help, About the Google Display Network,

GOOGLE, https://perma.cc/6582-YRSK (last visited Sept. 29, 2020). When taken to-

gether, the terms appear to permit Google to process bids that advertisers submit

via Google’s buying tool for small advertisers called Google Ads through two

different Google marketplaces (auctions). In other words, Google Ads hosts a

first auction, then Google Ads acts as the “buyer” in Google’s exchange, so that

Google simultaneously acts on the buy-side and the sell-side. In a recent submis-

sion to the Australian competition authority, Google implicitly confirms this

practice. See DANIEL S. BITTON & STEPHEN LEWIS, CLEARING UP MISCONCEPTIONS

ABOUT GOOGLE’S AD TECH BUSINESS 48 (May 5, 2020), https://perma.cc/WT2W-

DD74. At a 2019 conference for antitrust experts, Google chief economist Hal

Varian confirmed that Google can act on both the buy-side and the sell-side at

the same time and explained that Google uses a “formulaic apportionment” to

price ads when Google participates on both sides of a transaction. Stigler Center,

2019 Antitrust and Competition Conference, Pt. 11: Fireside Chat, YOUTUBE, at 1:01:13

(June 21, 2019), https://perma.cc/K2RR-UTCT. Finally, another point to consider

is that arbitrage, hidden fees, and undisclosed kickbacks may be a pervasive

problem in this electronic trading market. See Sarah Sluis, Investigation: DSPs

Charge Hidden Fees – And Many Can’t Afford to Stop, ADEXCHANGER (Jan. 10, 2018),

https://perma.cc/7GHG-KWDC.

. Websites that sell ads in Google’s exchange can see buyers’ clearing

prices via the centralized market reports that Google shares back with sites.

However, the buyers here are the intermediaries, such as Google Ads, not adver-

tisers.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

High trading costs in this market also affect consumers. As a general

matter, if publishers like The Post and The Register make less money sell-

ing ads, they have less to re-invest into the business of investigative jour-

nalism and news. But the news business globally is already struggling. In

the last decade and a half, this sector in the U.S. has shed 51% of news-

room jobs, paywalls and subscription prices have increased, and 20% of

newspapers have closed.

This contraction has led some economists to

urge lawmakers to think about democracy and “citizen welfare” when

considering competition problems in advertising.

Lawmakers, antitrust enforcers, and academics are concerned about

distorted growth and high trading costs, which ultimately harm con-

sumer welfare. As a result, they have been asking, why is competition not

working better? In the U.S. and globally, governments are investigating

whether Google has monopolized advertising markets or restrained com-

petition by engaging in specific conduct that violates competition laws.

. See generally Felix Simon & Lucas Graves, Across Seven Countries, the

Average Price for Paywalled News is About $15.75/Month, NIEMANLAB (May 8, 2019),

https://perma.cc/D8CF-CN46 (surveying rise of paywalls); Freddy Mayhew,

More Paywalls Going Up Online as News Publishers Face Shrinking Share of Ad Reve-

nue and Try to Fight Back Against Ad-Blockers, PRESSGAZETTE (May 15, 2018),

https://perma.cc/CHR2-SBJC; Elizabeth Grieco, U.S. Newspapers Have Shed Half of

Their Newsroom Employees Since 2008, PEW RSCH. CTR. (Apr. 20, 2020),

https://perma.cc/NN29-Z36V (“The number of newspaper newsroom employees

dropped by 51% between 2008 and 2019, from about 71,000 workers to 35,000.”);

PENELOPE MUSE ABERNATHY, THE EXPANDING NEWS DESERT (UNC Ctr. for Inno-

vation and Sustainability in Loc. Media ed., 2018) (finding total number of U.S.

newspapers declined from 8,891 in 2004 to 7,112 in 2018).

. STIGLER COMMITTEE ON DIGITAL PLATFORMS, FINAL REPORT (Stigler Ctr.

for the Study of the Economy and the State ed., 2019) [hereinafter STIGLER

COMMITTEE REPORT] (discussing how the news market in democratic societies

shares features of a public good and how either concentration in news outlets or

increased news prices harms “citizen welfare”).

. See Tony Romm, 50 U.S. States and Territories Announce Broad Antitrust

Investigation of Google, WASH. POST (Sept. 9, 2019), https://perma.cc/K8LA-NA2H;

Keach Hagey & Rob Copeland, Justice Department Ramps Up Google Probe, With

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

Do digital markets naturally tend to monopolize because of network ef-

fects?

Or has Google monopolized these markets by unlawfully exclud-

ing competition?

At the same time, economists and other scholars have

Heavy Focus on Ad Tools, WALL ST. J. (Feb. 5, 2020), https://perma.cc/CV82-T6K8

(reporting that the U.S. Justice Department's antitrust probe into Google is focus-

ing heavily on Google's advertising products); Silvia Amaro, EU Starts New Pre-

liminary Probe into Google and Facebook’s Use of Data, CNBC (Dec. 2, 2019),

https://perma.cc/BL8S-8WFU; Mark Sweney, Google and Facebook Under Scrutiny

over UK Ad Market Dominance, GUARDIAN (July 3, 2019), https://perma.cc/7EQR-

X7FC; Jamie Smyth, Australia Probes Impact of Facebook and Google on Media, FIN.

TIMES (Dec. 3, 2017), https://perma.cc/Z46B-PA6L.

. There has been push-back on the idea that network effects and econo-

mies of scale can explain market concentration in “big-tech” markets and a

broader conversation about market concentration problems in the United States.

See THOMAS PHILIPPON, THE GREAT REVERSAL 267 (Belknap Press of Harvard

Univ. Press, 2019). But see STIGLER COMMITTEE REPORT, supra note 18 (discussing

in broad terms tendency for digital platforms to concentrate markets due to net-

work effects and other factors); Lina M. Khan, The Separation of Platforms and Com-

merce, 119 COLUM. L. REV. 973 (2019) (observing generally that digital platforms

may tend to tip to monopolies and how network effects can act as a barrier to

entry).

. Competition laws in the U.S. and globally prohibit firms with market

power from engaging in conduct that is exclusionary. In the U.S., Sections 1 and

2 of the Sherman Act prohibit exclusionary conduct. 15 U.S.C. §§ 1-2 (1890); see

generally Herbert J. Hovenkamp, The Antitrust Standard for Unlawful Exclusionary

Conduct, 1777 FAC. SCHOLARSHIP AT PENN L. 1 (June 2008) (defining exclusionary

conduct as that “reasonably capable of creating, enlarging or prolonging monop-

oly power by impairing the opportunities of rivals,” where such conduct either

does “not benefit consumers at all,” is “unnecessary for the particular consumer

benefits claimed,” or “produce[s] harms disproportionate to benefits”); United

States v. Microsoft Corp., 253 F.3d 34, 49-50 (D.C. Cir. 2001) (discussing single

firm exclusionary conduct in technologically dynamic markets). In Europe, Arti-

cle 102 of the Treaty on the Functioning of the European Union (TFEU) similarly

prohibits firms from using their dominant position in a market to undermine

competition. See Guidance on Article 102 Enforcement Priorities, 2008 O.J. (C 115)

89. In Australia, Section 46 of the Trade Practices Act of 1974 prohibits a company

“in a position substantially to control a market” from leveraging that position to

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

been debating how to spur competition outside the scope of antitrust en-

forcement.

Some have advocated for the creation of a specialized digital

competition authority, while others have argued more generally for

structural separations.

This Article studies the structure of online advertising markets, what

drives competition, and how Google became dominant by engaging in

conduct that lawmakers prohibit in other electronic trading markets. To

“fix” competition in advertising, policymakers might lean on the toolbox

exclude competition. Trade Practices Act 1974 (Cth) s 46 (Austl.). In China, Arti-

cle 6 of China’s Anti-Monopoly Law prohibits dominant firms from leveraging

their position to “eliminate or restrict competition.” Fàn Lǒngduàn Fǎ (反垄断法

) [Anti-Monopoly Law of China] (promulgated by the Standing Comm. Nat’l

People’s Cong., Aug. 30, 2007, effective Aug. 1, 2008), art. 6.

. STIGLER COMMITTEE REPORT, supra note 18 (proposing interoperability,

stronger merger guidelines and antitrust enforcement, data remedies, and pro-

consumer default rules, amongst others); STIGLER CTR., PROTECTING JOURNALISM

IN THE AGE OF DIGITAL PLATFORMS (2019), https://perma.cc/9LPT-

AQB4 (proposing public funding of news operations, and a requirement that

Google and other digital companies prioritize content according to criteria other

than ad revenue, amongst other measures); CMA FINAL REPORT, supra note 7

(considering regulating dominant digital platforms with a code of conduct);

Madhumita Murgia & Kate Beioley, UK to Create Regulator to Police Big Tech Com-

panies, FIN. TIMES (Dec. 18, 2019), https://perma.cc/DE8S-EXYM (reporting that

the UK in 2020 will move forward with establishing a new tech regulator for

companies such as Google and Facebook); RICHARD KRAMER, CMA ONLINE

PLATFORMS REVIEW: ARETE RESEARCH’S VIEW (2019), https://perma.cc/2F7J-W4LM

(drawing parallels between permitted conduct in advertising markets and pro-

hibited conduct in financial markets). See JASON FURMAN ET AL., UNLOCKING

DIGITAL COMPETITION: REPORT OF THE DIGITAL COMPETITION EXPERT PANEL (2019);

JACQUES CRÉMER ET AL., COMPETITION POLICY FOR THE DIGITAL ERA (Eur. Comm’n

ed., 2019).

. See generally supra note 19; Khan, supra note 20 (arguing that lawmakers

should consider prohibiting dominant “digital platforms” from both running a

market and participating in it and that structural separations would be more de-

sirable than non-discrimination rules that would require case-by-case adjudica-

tion); Elizabeth Warren, Here’s How We Can Break Up Big Tech, MEDIUM (Mar. 8,

2019), https://perma.cc/2SYT-T7TX (proposing separating platforms and plat-

form participants and/or fair and non-discriminatory rules of dealing).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

that financial regulators have already developed to protect the integrity

of a parallel real-time trading market, the securities market. That toolbox

provides a framework for understanding and addressing competition

problems in advertising.

Part II begins this conversation by identifying segments of the econ-

omy that have migrated to electronic trading, discussing the structure of

these markets, and explaining how the structure of online advertising

markets, in which Google is dominant, is similar. In markets with this

structure, problems related to concentration and distorted growth can re-

sult when exchanges provide a subset of traders with information or

speed advantages. Problems can also result when trading intermediaries

route orders (i.e., liquidity) to an exchange in a discriminatory manner or

abuse their access to third parties’ sensitive nonpublic information. In the

stock market, lawmakers safeguard competition by requiring exchanges

to give all traders non-discriminatory access to the marketplace, by iden-

tifying and managing intermediary conflicts of interest, and by requiring

trading disclosures to advance both principles. The integrity of the ad-

vertising market does not benefit from parallel competition safeguards.

Part III examines Google’s extensive conflicts of interest and specific

conduct in advertising. Part A first discusses how the story of Google’s

rise is in part a story of information and speed asymmetry: Google’s ex-

change advantages the Google-owned intermediaries with better infor-

mation about the ad space trading on Google’s exchange and with speed

advantages. When it comes to information asymmetries, Google often

weds its practice of cutting off rivals’ access to data in the noble language

of furthering user privacy. But users’ privacy is not protected from

Google, only from trading rivals, and competition is stunted as a result.

Part B then explains how Google routes buy and sell orders in a discrim-

inatory manner, to both its exchange and web properties. Finally, Part C

considers how Google distorts competition by using the sensitive non-

public information belonging to third-party buyers and sellers—infor-

mation it becomes privy to as an intermediary—to inform its own trading

activity in the market.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

Although Google’s conduct may appear novel and unprecedented,

Part IV discusses in more depth how policymakers have dealt with par-

allel conduct in other electronic trading markets. As additional segments

of the economy have migrated to real-time electronic trading, including

small and emerging sectors like event tickets, airfare tickets, and crypto-

currencies, lawmakers have monitored for these common competition

problems, and they have at times intervened with legislation or regula-

tion when market participants have not sufficiently self-regulated their

behavior. Because the advertising sector now trades on electronic ex-

changes too, we might similarly borrow from the principles of financial

regulation—ensuring equal access to speed and information and manag-

ing intermediary conflicts of interest—in order to protect the integrity of

advertising.

II. ELECTRONIC TRADING MARKETS

A. Advertising Market Reflects the Structure of an Electronic Trading

Market

The rise of buying and selling ad space on electronic exchanges par-

alleled a shift to computerized trading systems across various sectors of

the economy. For example, shares of issued stock trade on stock ex-

changes like the New York Stock Exchange (NYSE), which once reflected

a hustle and bustle of traders on a physical exchange floor. Though tele-

vision networks like CNBC still perpetuate this image, today, stocks, cur-

rencies, and other financial instruments trade on dozens of electronic

trading venues at the same time and at lighting speed.

The NYSE is

. For a discussion of how equities trading happens on national public

exchanges, off-exchange alternative trading systems, and internalization plat-

forms, see Merritt Fox et al., The New Stock Market: Sense and Nonsense, 65 DUKE

L.J. 191 (2015); Kevin S. Haeberle, Discrimination Platforms, 42 J. CORP. L. 809

(2017); MATTEO AQUILINA ET AL., QUANTIFYING THE HIGH-FREQUENCY TRADING

“ARMS RACE”: A NEW METHODOLOGY AND ESTIMATES (Fin. Conduct Auth. ed.,

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

largely comprised of computer servers, interconnected by colored wiring

in nondescript buildings off the New Jersey Turnpike.

To buy and sell on these financial exchanges, investors used to go

through a human middleman. But now brokers and other intermediaries

are also computerized, connecting electronically to exchanges through

application programming interfaces. An individual investor might use

an online interface belonging to a broker like E*Trade, while an institu-

tional trader might use sophisticated algorithms to trade at high speeds

in an automated fashion.

Outside of tradeable financial assets like equities, the tickets for

sports, theater, and music events now trade on electronic marketplaces

and “exchanges,” as do emerging cryptocurrencies like Bitcoin and

Ether.

To buy and sell on these electronic trading venues, one can use

an online interface or go through a “broker.” Here too, trading strategies

can make use of computerized algorithms to buy and sell in an auto-

mated way at high speeds.

The biggest financial players have been helping to propel electronic

trading to new sectors of the economy. For example, the parent company

of the NYSE, Intercontinental Exchange, recently made a takeover bid for

eBay (withdrawn), which, until not long ago, also owned the largest

event ticket marketplace, StubHub.

Separately, large financial brokers,

2020); SCOTT PATTERSON, DARK POOLS: THE RISE OF THE MACHINE TRADERS AND

THE RIGGING OF THE U.S. STOCK MARKET (Crown Bus. ed., 2013).

. Graham Bowley, The New Speed of Money, Reshaping Markets, N.Y. TIMES

(Jan. 1, 2011), https://perma.cc/4VEB-F987.

. Note, tickets trade in primary and secondary markets. See generally U.S.

GOV’T ACCOUNTABILITY OFF., GAO-19-347, EVENT TICKET SALES: MARKET

CHARACTERISTICS AND CONSUMER PROTECTION ISSUES (2018); ERIC T.

SCHNEIDERMAN, OBSTRUCTED VIEW: WHAT’S BLOCKING NEW YORKERS FROM

GETTING TICKETS (N.Y. State O.A.G. ed., 2016); BARBARA UNDERWOOD, VIRTUAL

MARKETS INTEGRITY INITIATIVE REPORT (N.Y. State O.A.G. ed., 2018) (providing an

overview of cryptocurrency trading).

. Cara Lombardo & Corrie Driebusch, NYSE Owner Intercontinental Ex-

change Makes Takeover Offer for EBay, WALL ST. J. (Feb. 4, 2020),

https://perma.cc/79DL-SW38.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

including E*Trade and TD Ameritrade, are launching cryptocurrency ex-

changes and trading desks.

NYSE rival, Nasdaq, has taken a different

approach, licensing its underlying marketplace technology to jumpstart

other sectors’ migration to electronic trading.

One exchange built on the

Nasdaq framework is NYIAX, the New York Interactive Advertising Ex-

change.

But the biggest names in the advertising market are not NYSE or

Nasdaq; they are Google and Facebook. These companies, amongst the

largest market cap companies today, operate the nuts and bolts of what

is likely the most sophisticated of all electronic trading markets: online

advertising. Whereas the world’s largest financial exchange, the NYSE,

trades the shares of a few thousand companies and processes a few bil-

lion shares a day, Google’s advertising exchange trades ad spaces tar-

geted to billions of individual users and likely processes tens of billions

of these targeted ad spaces daily.

Just as individual investors go through an intermediary broker to

trade on financial exchanges, publishers and advertisers must also go

through a computerized middleman to trade on advertising exchanges.

On the buy-side, advertisers use specialized software made either for

small or large advertisers. Smaller advertisers, such as your local dry

. Elizabeth Dilts, TD Ameritrade Invests in Cryptocurrency Exchange ErisX,

REUTERS (Oct. 3, 2018), https://perma.cc/KPE7-RZEL; Julie Verhage, E*Trade Is

Close to Launching Cryptocurrency Trading, BLOOMBERG (Apr. 26, 2019),

https://perma.cc/SE4K-H888.

. Non-Traditional Exchanges & New Markets, NASDAQ,

https://perma.cc/F4TN-N4DM (last visited Sept. 29, 2020).

. Announcing NYIAX, the World’s First Advertising Contract Exchange,

NASDAQ (Mar. 14, 2017), https://perma.cc/8UZT-FDS7.

. Jeff Desjardins, Here’s the Difference Between the NASDAQ and NYSE,

BUS. INSIDER (July 11, 2017), https://perma.cc/L6YQ-BF77; Markets Diary, WALL ST.

J.: MKTS., https://perma.cc/F7SV-LTQC (last visited Sept. 29, 2020) (reporting

NYSE daily trading volume); Chiradeep BasuMallick, What Is an Ad Exchange?

Definition, Functioning, Types, and Examples, TOOLBOX: MKTG. (May 18, 2020),

https://perma.cc/9AW7-SN9G (estimating that advertising exchanges process

approximately 70 billion ad “impressions” (i.e., ad spaces) daily).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

cleaner, might use a simple, self-serve online buying tool, such as Google

Ads, which Google has analogized to the “online broker” in the ad mar-

ket.

Continuing this analogy, an early Google Ads competitor called it-

self “the eTrade to Google’s NYSE.”

In practical terms, the dry cleaner

sets a budget and defines its bid parameters (e.g., bid ceiling) and Google

Ads will bid on and buy ad space, including those trading on Google’s

exchange, in an automated fashion on the dry cleaner’s behalf.

How-

ever, Google here can ultimately be the advertiser’s counterparty, not its

agent.

Larger advertisers like Proctor and Gamble use enterprise trading

software that the industry calls demand side platforms (DSPs) and, again

. Comparisons between the advertising market and the financial market

have frequently been made by advertising industry participants, including

Google and others. For instance, Google compared Google Ads (then called Ad-

Words) to an online broker: “Who participates in the Ad Exchange? Again, im-

agine the Ad Exchange as a stock exchange. Only the largest brokerage houses

actually plug into, say, the NYSE. In the Ad Exchange world, those are: The large

online publishers (sellers)—websites like portals, entertainment sites and news

sites Ad networks and agency holding companies that operate networks (buy-

ers)—companies that connect web sites with advertisers.” The DoubleClick Ad Ex-

change, GOOGLE, https://perma.cc/5TTF-7PEK (last visited Oct. 3, 2020); see also

Mansour et al., supra note 4 (analogizing “like financial exchanges that only let

licensed brokers trade, ad exchanges let ad networks trade on the exchange on

behalf of individual advertisers.”).

. AdExchanger Staff, On DoubleClick Ad Exchange: More Digital Media In-

dustry Reaction, ADEXCHANGER (Sept. 22, 2009), https://perma.cc/75GY-D87J

(comments of then-CEO of buying tool AdReady).

. Note, historically, Google Ads only routed bids to Google’s exchange.

This changed in 2016 when Google Ads started routing bids to non-Google ex-

changes.See DoubleClick Ad Exchange, DOUBLECLICK BY GOOGLE,

https://perma.cc/TZ6F-S55S (last visited Oct. 3, 2020) (explaining that Google’s

exchange is “the only exchange offering access to the full demand of Google Ad-

Words”); Michel Van Luijtelaar, AdWords Remarketing Cross-Exchange, UP

ANALYTICS (Aug. 26, 2016), https://perma.cc/34GY-WJ2P (reflecting that Google

Ads/AdWords started routing bids to non-Google exchanges in May of 2016);

Google Ads Help, About Cross-Exchange for Display Remarketing Campaigns,

GOOGLE, https://perma.cc/NDP3-TDUE (last visited Sept. 29, 2020).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

borrowing from finance, “trading desks.”

Compared to the self-serve

software used by smaller advertisers, these software tools provide more

sophisticated bidding algorithms, offer a wider array of user targeting

options, and typically require a higher monthly spending commitment.

For brevity, this Article refers to self-serve tools for small businesses,

DSPs, and trading desks, collectively, as “buying tools.”

. See generally Michael Sweeney, The Anatomy of a Demand-Side Platform

(DSP), CLEARCODE, https://perma.cc/K9EQ-XKBH. Note, demand side platforms

and trading desks also connect to the ad serving tool used by marketers. See gen-

erally Arvind Kesh, What Is a Demand-Side Platform and How to Choose One,

ADBEAT, https://perma.cc/K86U-J2UK.

. See generally Kesh, supra note 35 (summarizing that most DSPs require

monthly spend commitments of $5,000-$10,000 U.S. dollars).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

Figure 1: The Buy-Side Intermediaries: Buying Tools*

*Advertisers must use buying tools to access the exchanges where ads

are bought and sold.

The counterparts of advertisers in this market are those selling ad

space, including publishers such as The Post and The Register. Sellers use

a different type of computerized intermediary called an “ad server” to

sell their inventory on exchanges.

At a simple level, the ad server is in-

ventory management software, which keeps track of the number of ad

spaces a publisher has available to sell and houses sensitive information

about the publisher’s campaigns, advertisers, and pricing.

As a part of

. It is worth noting that Google recently blurred the distinction between

its ad server and exchange by both reclassifying its ad serving revenues in its

shareholder reports and merging the two into a new single product renamed

Google Ad Manager (GAM). Jonathan Bellack, Introducing Google Ad Manager,

GOOGLE: AD MANAGER (June 27, 2018), https://perma.cc/T3HN-YCZV (announc-

ing that Google has merged its ad server and exchange together and renamed

them Google Ad Manager). However, from 2008 through 2018, Google’s ad

server and exchange were marketed as separate, distinct, products. Additionally,

in every 10K SEC filing from 2008 (the year Google acquired DoubleClick)

through 2014, Google distinguished its “ad serving software” from its exchange

and ad network. It was only in 2015 that Google reclassified its ad serving soft-

ware revenues from “other” to “advertising revenues” and stopped referring to

its ad server as a separate product. See Google Inc. (2008-2015), supra note 10.

. NetGravity Launches AdServer, the Premier Advertising Management Sys-

tem Software for World Wide Web Publishers, NETGRAVITY (Jan. 31, 1996),

https://perma.cc/7B2W-JB4M (announcing launch of first “adserver”); Julia

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

this role, the ad server also acts as a link between a publisher’s inventory

and real-time trading venues, routing ad space to exchanges in real-time

as they become available for sale.

After a publisher sells its ad space,

information about the order goes into the ad server. The ad server then

tracks these orders and ensures each advertiser’s ads are displayed (i.e.,

are served) in the right spot, at the right time, to the right users.

Figure 2: The Sell-Side Intermediary: Ad Server*

Brockhoff et al., Google/DoubleClick: The First Test for the Commission’s Non-Hori-

zontal Merger Guidelines, 2 COMPETITION POL’Y NEWSL. (Eur. Comm’n Competi-

tion), 2008, at 53 (explaining the core functions of an ad server); DoubleClick Inc.

and Compaq Computer Corp., Advertising Services Agreement, FINDLAW (Jan. 18,

1999), https://perma.cc/5T83-RGJ3 (detailing commercial terms of an old Dou-

bleClick license).

. Specifically, when a user loads a publisher’s webpage, the user loads

the ad server’s “tag,” which is commonly Google’s since Google owns the lead-

ing ad server in the market. Historically, Google’s ad server limited interconnec-

tion with non-Google exchanges. See infra Part III.B. When this was the case,

Google’s ad server redirected the user’s browser to call non-Google exchanges

directly. See How RTB Ad Serving Works, AD OPS INSIDER (Dec. 15, 2010),

https://perma.cc/S3TR-T84E (discussing how it worked historically). This process

changed in part when Google’s ad server introduced a product enhancement

called Open Bidding. With Open Bidding, it is Google’s ad server (and not the

user’s browser) that makes direct calls to integrated exchanges. See Jonathan Bel-

lack, Improving Yield, Speed and Control with DoubleClick for Publishers First Look

and Exchange Bidding, GOOGLE AD MANAGER (Apr. 13, 2016),

https://perma.cc/LMU4-ENCQ.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

*Websites must use an ad server to access the exchanges where ads are

bought and sold.

The lifecycle of an ad trade flows through these three software com-

ponents—the ad server, the exchanges, and the buying tools—and begins

the moment a user visits a webpage.

The user’s visit triggers the pub-

lisher’s ad server to identify the user loading the page and to route the ad

space on that page to one or more exchanges. The exchange then sends

trading signals called “bid requests” to the buying tools that have a “seat”

to bid, soliciting them to return a bid for that space without knowing

what others are simultaneously returning as their bid.

Each exchange

then holds an auction, picks a winning bid, and returns it to the ad server.

The ad server can then maximize the publisher’s inventory yield by se-

lecting the advertisement associated with the highest exchange bid and

returning it to the user’s page all before it finishes loading.

. Websites might use another software called a supply side platform

(SSP) between their ad server and exchanges. The SSP’s job is to route the site’s

ads to exchanges in a way that maximizes yield. However, the ability today to

route ad space to multiple exchanges synchronously largely renders SSPs obso-

lete, which is why the delineation between SSPs and exchanges has largely dis-

appeared. See Maciej Zawadziński & Michal Wlosik, What is a Supply-Side Plat-

form (SSP) and How Does It Work?, CLEARCODE (Oct. 18, 2018),

https://perma.cc/F5TG-UKVA (explaining what an SSP is and does); Ryan Joe,

Defining SSPs, Ad Exchanges and Rubicon Project, ADEXCHANGER (Feb. 7, 2014),

https://perma.cc/FQP2-JA8Q (describing the disappearing delineation between

SSPs and exchanges).

. See generally Verizon Media, Auction Mechanics 101, VERIZON (Jan. 30,

2019), https://perma.cc/53XX-CAQU (explaining how advertising auctions

work); Authorized Buyers Real-Time Bidding Proto, GOOGLE,

https://perma.cc/YKE6-N6DX (explaining how Google shares advertisers’ bids

post auction conclusion).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

Figure 3: Electronic Ad Trading Altogether

B. Common Competition Problems in Electronic Trading Markets

In the online advertising market, as well as in other electronic trading

markets, access to information, speed, and the routing of buy and sell

orders are the linchpin of a healthy, competitive market. Access to infor-

mation about what is trading on an exchange is critical to those compet-

ing to buy on the same venue. In advertising, the bid requests that ex-

changes send to buying tools contain important information used to

decide whether and how much to bid for an ad. This includes the size of

the ad space for sale (e.g., 300x250 pixels), the page address (e.g., ny-

times.com/HowToDoLaundry), and some information about the identity

of the user.

Importantly, when these bid requests do not contain suffi-

cient information about the identity of the user loading the page, which

people in the industry have called the “skeleton key” of programmatic

. For a list of what Google includes in bid requests, see Google Develop-

ers, Real-Time Bidding: Example Bid Request, GOOGLE, https://perma.cc/X4ND-

77V9 (last visited Oct. 4, 2020). For an example of what another major exchange

includes in bid requests, see Xandr Bidders, Open RTB 2.0 Bid Request, XANDR,

https://perma.cc/AW5X-ST22 (last visited Sept. 29, 2020), which explains bid re-

quests and the fact that they contain “all the necessary information for a bidder

to produce a bid price.”

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

advertising, the buying tools bidding on behalf of advertisers sit out of

auctions or bid significantly less.

Speed is critical to electronic trading, whether on advertising, stock,

ticket, or cryptocurrency exchanges.

Online ad space trades in the mil-

liseconds that it takes for users’ pages to load and exchanges and buying

tools communicate with each other at lightning speed. When an adver-

tising exchange sends out bid requests, it sets the time each buying tool

has to respond with a bid. Within this timeframe, which is usually be-

tween 100 to 160 milliseconds (one to two-tenths of a second), each tool

races to unpack the data contained in the bid request, query additional

user data (e.g., this particular user’s spending habits), determine what

price to bid, and return a bid back to the exchange before time is up.

. Andrew Casale, Identity: Programmatic’s Skeleton Key, VIMEO, at 5:40

(May 21, 2018), https://perma.cc/E28F-EXQJ.

. See generally FirstPartner, Digital Advertising: The Role of Cloud and Con-

nectivity in Ad Trading and Delivery, INTERXION (2020), https://perma.cc/67ZR-

QTUT (explaining that “very fast response times” are “a precondition for com-

peting” and “[t]rading and delivering ads at very high speed is a critical require-

ment, and depends on rapid interactions between partner companies”); Hasham,

How Network Latency Affects the RTB Process for AdTech, DATAPATH (Apr. 21, 2016),

https://perma.cc/4W6Q-GUL2 (explaining that a “small improvement in latency

can spell the difference between winning an auction and not being considered

for the auction at all”); Tejaswini Tilak, NEED FOR SPEED: Why the Online Ad

Industry Is Converging on Equinix, EQUINIX (Nov. 18, 2013),

https://perma.cc/G6GE-TV48.

. Google Ad Manager, Bring More Bids to the Auction with Open Bidding,

GOOGLE, https://perma.cc/YPH2-KPAL (last visited Sept. 29, 2020) (sharing new

Google ad server Open Bidding timeouts of 160 milliseconds); BidResponse Object,

OPENX (Oct. 9, 2017), https://perma.cc/37ME-6S9E (discussing OpenX exchange

timeouts of 125 milliseconds and how the exchange reduces the number of bid

requests sent to bidders with frequent timeouts); PubMatic Technical Documen-

tation, OpenRTB 2.1 API Performance, PUBMATIC, https://perma.cc/36BA-42PK

(last visited Sept. 29, 2020) (“PubMatic’s performance requires that the total la-

tency should be within 130 milliseconds - 30 ms for connection establishment and

100 ms for bid response. If either of these independent thresholds are exceeded

(connection and bid response time) during the transaction the bid is considered

a ‘timeout.’”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

After the set time, the exchange closes the auction, excludes the bids that

arrived too late, and chooses a winner.

Across any of these electronically traded markets, an exchange can

distort competition between the different buyers competing in its mar-

ketplace by giving some an information or speed advantage. In advertis-

ing, an exchange might give some buying tools (bidders) superior infor-

mation about the ad space (e.g., the user’s identity) or let some colocate.

Colocation, which is also central to equities trading, broadly refers to the

practice of placing trading computers and exchange computers close to-

gether to reduce the time it takes for signals to travel between the two.

By colocating with an ad exchange, a bidder can receive and respond to

bid requests faster than the bidders that are not colocated, allowing it to

be included more often in exchanges that subject their auctions to strict

time constraints.

Just as exchanges can distort competition between bidders, the trad-

ing intermediaries can distort competition between exchanges by the way

that they route buy and sell orders to exchanges. When ad space on a site

becomes available for sale, the ad server—like the broker in financial

markets—determines whether to route that space only into Exchange A,

or Exchange A, B, and C, on equal terms, and whether to do so at the

same time. Similarly, when an advertiser uses a buying tool to bid on and

buy ad space from exchanges, this intermediary determines whether to

route the advertiser’s bids only into Exchange A, or Exchange A, B, and

C, on equal terms. A company that operates an intermediary, especially

one that has significant market share and enjoys barriers to entry, can

distort competition in the exchange market by, for example, preferen-

tially routing buy and sell orders to a particular trading venue.

. Geoffrey Rogow, Colocation: The Root of All High-Frequency Trading

Evil?, WALL ST. J. (Sept. 20, 2012), https://perma.cc/QM7B-HSA3 (defining colo-

cation); Google Ad Manager Help, How Google Ads and Display & Video 360 Work

with Ad Exchange, GOOGLE, https://perma.cc/MWY3-JLP5 (last visited Sept. 26,

2020) (discussing speed and other advantages of colocation).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

Also critical to competition is the way that trading intermediaries

handle the material nonpublic information belonging to third party buy-

ers and sellers. Problems often arise when companies trade on behalf of

third parties, but also trade on behalf of themselves. For example, in fi-

nancial markets, a broker might receive a Carl Icahn sell order for 10 mil-

lion shares of Tesla. The broker can best serve Icahn’s interests by keeping

information about his trading activity confidential, or alternatively, the

broker can use that information to advance its own interests. For instance,

its proprietary trading division might use information about Icahn’s

trade to get rid of its Tesla shares before information about Icahn’s trade

becomes public and the price of those shares drops. To be discussed more

in Part III, information use problems also arise in advertising markets

when a company both handles trading activity for third parties but also

buys and sells in the market for its own financial interests.

C. The Competition Principles We Apply in the Equities Trading

Market

In the U.S. and globally, lawmakers manage these common electronic

trading issues in the stock market through the application of a handful of

broad principles.

While financial market regulation may sound intimi-

dating, the basic principles are straightforward. One guiding principle is

that exchanges must provide traders with fair access to the marketplace,

including access to the data transmitted by exchanges as well as the speed

at which data signals travel from exchanges to traders.

When exchanges

. U.S. Congress started regulating the stock market in 1934 to prohibit

unfair trading practices and “insure the maintenance of fair and honest markets.”

Securities Exchange Act of 1934, 15 U.S.C. § 78b.

. In the U.S., stock exchanges must obtain approval for their trading

rules from the SEC, who in turn must ensure “fair competition among brokers

and dealers.” 15 U.S.C. § 78k-1(a)(1)(C)(ii). Specific fair access rules are articu-

lated in federal statutes. For example, Rule 610(a) and Rule 603(a)(2) of SEC Reg-

ulation NMS, prohibit the regulated National Market System (NMS) exchanges,

including the NYSE and Nasdaq, from restricting efficient exchange access; these

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

permit colocation, terms must be transparent, pricing non-discrimina-

tory, and the length of fiber-optic cord connecting the exchange engine to

the trader servers the same length.

Even an extra foot of cabling in a

colocation facility can systemically disadvantage some traders due to la-

tency.

Competition is also protected through the identification and manage-

ment of intermediary conflicts of interest.

One principle applied here is

rules also require exchanges to distribute information with respect to quotations

or transactions in a manner “fair and reasonable” or “not unreasonably discrim-

inatory.” See respectively 17 C.F.R. § 242.610(a) (2020) (prohibiting restrictions on

“efficient access”); 17 C.F.R. § 242.603(a)(2) (2020). The SEC has interpreted these

fair access rules to prohibit an exchange from sending data to some traders before

the exchange sends the same data to consolidated feeds. See, e.g., N.Y. Stock Exch.

LLC, Release No. 67857 SEC, 2 (Sept. 14, 2012). While the non-NMS exchanges—

the Alternative Trading Systems (ATSs)—are generally not subject to fair access

rules, some are subject to the fair access standards of Regulation ATS. 17 C.F.R.

§ 242.301 (2020). In the European Union, access to regulated markets (RMs) and

multilateral trading facilities must be transparent and non-discriminatory. Other

jurisdictions globally also require trading venues to provide traders with fair ac-

cess. See generally INT’L ORG. OF SECURITIES COMM’NS [IOSCO], REGULATORY ISSUES

RAISED BY CHANGES IN MARKET STRUCTURES CONSULTATION REPORT 9 (2013) [here-

inafter IOSCO MARKET STRUCTURE CONSULTATION REPORT].

. Exchange colocation procedures need to be submitted to and approved

by the SEC. For transparency around exchange colocation services, see, e.g., NYSE

Price List 2020, N.Y.S.E. 28-39 (2020), https://perma.cc/T2XU-HFDF; MORGAN

HAUSEL, THE BAD SIDE OF A GOOD IDEA (Collaborative Fund ed., 2016),

https://perma.cc/AX7A-M6V6 (“NYSE measured the distance to the furthest cab-

inet, which is where people put their servers. It was 185 yards. So they gave every

[high-frequency trader] a cable of 185 yards.”)

. See generally Resolution on IOSCO Objectives and Principles of Securi-

ties Regulation and Methodology for Assessing Implementation of the IOSCO

Objectives and Principles of Securities Regulation, Presidents Comm. of the Int’l

Org. of Securities Comm’ns [IOSCO] (May 2017), https://perma.cc/UG65-KS23

(summarizing that one key role of the securities regulator, or an industry self-

regulatory organization, is to avoid, eliminate, disclose, or otherwise manage

conflicts of interest); Carlo V. di Florio, Conflicts of Interest and Risk Governance,

S.E.C. (Oct. 22, 2012), https://perma.cc/L3RW-NY9W (providing an overview of

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

a structural one: a company that runs an exchange like the NYSE cannot

also operate a division involved in trading.

However, when companies

are permitted to participate in the market in more ways than one, conflicts

of interest and disclosure rules kick in. For example, intermediary broker

dealers can have conflicts: they trade in the market on behalf of third par-

ties (the broker designation), as well as on behalf of themselves as a pro-

prietary trader (the dealer designation), and they can even run a special-

ized trading venue called an Alternative Trading System (ATS).

But

these multi-service firms must manage their conflicts of interest and can-

not simply route their customers’ buy and sell orders (order flow) to the

firm’s electronic trading venue.

the importance of conflicts of interest management to securities regulation and

the interplay between the existence of conflicts and increased market risk); Chris-

toph Kumpan & Patrick C. Leyens, Conflicts of Interest of Financial Intermediaries -

Towards a Global Common Core in Conflicts of Interest Regulation, 4 EUR. CO. AND

FIN. L. REV. 72 (2008) (defining a conflict of interest as arising when “a person

who has a duty to act in another party’s interest has to decide how to act in the

interest of that party and another interest interferes with his ability to decide ac-

cording to his duty.”).

. Conversations with securities professionals indicate that the SEC en-

forces this structural separation through its power to reject public exchange ap-

plications.

. Prop. Regulation of NMS Stock Alternative Trading Systems, 80 Fed.

Reg. 80,998 (Dec. 28, 2015) (codified at 17 CFR § 240.3a1-1(a)) (stating that broker

dealers operate ATSs).

. It is the Best Execution Rule, grounded in common law principles of

agency, industry self-regulation, and federal securities law, that prohibits de-

facto preferencing or internalization and requires broker dealers to use reasona-

ble diligence in determining where to route client orders. The rule’s application

in equities, however, is not uncontroversial and has application challenges (e.g.,

When does speed of execution trump best price? Is it unrealistic to apply such a

rule to individual trades?). See Jonathan Macey & Maureen O’Hara, The Law and

Economics of Best Execution, 6 J. OF FIN. INTERMEDIATION 188, 188-223 (1997). See

also Paul G. Mahoney & Gabriel V. Rauterberg, The Regulation of Trading Markets:

A Survey and Evaluation, in SECURITIES MARKET ISSUES FOR THE 21ST CENTURY, 221,

221-81 (Merritt B. Fox et al. ed., 2018).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

Competition in the stock market—heavily shaped by access to data

and information—also benefits from rules that regulate who may use

what information when trading. For example, we require financial inter-

mediaries (i.e., brokers and investment advisors) to act in the best interest

of their customers, which forcibly vests property rights in customers’ in-

formation in customers themselves.

Brokers, therefore, cannot use in-

formation about their customers’ trading activity to trade in the market

for their own financial gain.

This prohibition is further enforced through

. Regulation Best Interest under the Securities Exchange Act of 1934 re-

quires broker dealers to act in the “best interest” of their customers and not place

their own interest ahead of retail investors when advising on securities or invest-

ment strategy. This code of conduct requires broker dealers to mitigate or elimi-

nate certain some conflicts of interest and disclose others. It also requires broker

dealers to exercise reasonable diligence, care and skill when advising retail cli-

ents. Sec. Exch. Act of 1934, 17 CFR § 240.15l-1(a)(2) (2019). Investment advisors,

on the other hand, owe their customers a fiduciary duty, under the Investment

Advisors Act of 1940, 17 CFR § 275 (1940). The SEC has interpreted this duty to

permit investment advisors to sometimes merely disclose some types of conflicts

to institutional clients; to retail customers, investment advisors may instead have

to mitigate or eliminate conflicts entirely, especially complex ones. The fiduciary

standard for investment advisors also includes an ongoing duty to monitor cus-

tomer accounts. As background, per the Dodd-Frank Act, Dodd-Frank Wall

Street Reform and Consumer Protection Act, Pub. L. No. 111-203, § 929-Z, 124

Stat. 1376, 1871 (2010) (codified at 15 U.S.C. § 78o), the SEC was to “harmonize”

the duties of broker dealers and investment advisors. When promulgating a “best

interest” rule, as opposed to a parallel fiduciary rule, seven states, the District of

Columbia, and others, filed suit, challenging the short falling of the “best inter-

est” rule. Keith Blackman et al., Second Circuit Upholds Regulation BI, NAT’L L. REV.

(July 3, 2020), https://perma.cc/B8CY-D4TV.

. If a broker uses information about a customer’s trading activity to place

and execute trades in advance of the customer’s, it is called trading ahead or front

running, and is prohibited under common law principles of agency (which the

law constructs regardless of contractual intentions), industry self-regulation, and

federal securities law. See Opper v. Hancock Sec. Corp., 250 F. Supp. 668, 676

(S.D.N.Y. 1966) (holding front running to be illegal under principles of agency

and federal law), aff’d, 367 F.2d 157 (2d Cir. 1966); FINRA Rule 5270, FINRA

Rules, FIN. INDUS. REG. AUTH., https://perma.cc/5WPC-LMEF; NORMAN S. POSER,

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

Congressionally mandated ethical walls—internal corporate policies and

physical barriers that brokers must implement to prevent the flow of sen-

sitive information from one business division (e.g., the broker division)

to another (e.g., the dealer division).

On top of this, rules against insider trading further restrict certain

parties from trading upon particular types of information advantages.

The classic example prohibits corporate “insiders” from trading in the

market using material nonpublic information.

However, insider trading

BROKER-DEALER LAW AND REGULATION 16-5 (4th ed. 2007) (discussing broker

dealer duties).

. In 1988, Congress passed the Insider Trading and Securities Fraud En-

forcement Act of 1988, which increased penalties for insider trading and securi-

ties fraud and required registered brokers and dealers to enforce written policies

and procedures to prevent the misuse of material nonpublic information. See In-

sider Trading and Securities Fraud Enforcement Act of 1988, H.R. 5133, 100th

Cong. (1998). In 1990, the SEC clarified and outlined minimum requirements for

broker dealers. Section 204 of the Investment Advisers Act contains a similar re-

quirement for investment advisers. See 15 U.S.C. § 78o(g) (2018); DIVISION OF

MARKET REGULATION U.S. SEC. AND EXCH. COMM’N, BROKER-DEALER POLICIES AND

PROCEDURES DESIGNED TO SEGMENT THE FLOW AND PREVENT THE MISUSE OF

MATERIAL NONPUBLIC INFORMATION (U.S. Sec. and Exch. Comm’n ed., 1990); 15

U.S.C. § 80b-4a; RALPH C. FERRARA, DONNA NAGY & HERBERT THOMAS, FERRARA

ON INSIDER TRADING & THE WALL 9-7-9-12 (2017).

. Before the SEC took an active role in defining, expanding, and regulat-

ing “insider trading,” information use when trading was primarily governed by

state common law of fraud. The majority rule there rejected any fiduciary rela-

tionship between corporate insiders and shareholders, which would have trig-

gered a duty to disclose material nonpublic information to shareholders before

trading or to withhold from trading. At the helm of the SEC, Chairman William

Cary sought to reign in how certain parties with access to nonpublic information,

including corporate insiders and exchange floor members, could take advantage

of such information access. JOEL SELIGMAN, THE TRANSFORMATION OF WALL

STREET: A HISTORY OF THE SECURITIES AND EXCHANGE COMMISSION AND MODERN

CORPORATE FINANCE 344-45 (3rd ed. 1977). It was during Cary’s tenure that the

SEC began to use Section 10(b) of the 1934 Securities Exchange Act to prosecute

trading on the basis of certain information advantages. Though insider trading

doctrine developed over the years as requiring a breach of a fiduciary duty, some

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

rules also apply to “temporary insiders,” to prohibit those that become

privy to sensitive information through an unrelated business relationship

from trading on the same information.

These rules can promote eco-

nomic efficiency by properly allocating property rights to information

and by ferreting out trading activity thought to be grossly unfair.

In one

of the seminal insider trading cases at the U.S. Supreme Court, United

States v. O’Hagan, Justice Ginsburg relied on such rationale: unerodable

information advantages that derive from misappropriated information

can discourage people from participating in the markets in the first

place.

scholars have argued that such fiduciary duties were “fictions” manufactured to

prohibit trading based on information advantages deemed unfair. See, e.g.,

Donna Nagy, Insider Trading and the Gradual Demise of Fiduciary Principles, 94

IOWA L. REV. 1315 (2009) (discussing the “Supreme Court’s Fiduciary Fictions”).

. The development of insider trading rules stem from Rule 10b-5 of Sec-

tion 10(b) of the Securities Exchange Act (1934) that makes it unlawful “(a) to

employ any device, scheme, or artifice to defraud, (b) to make any untrue state-

ment of a material fact or to omit to state a material fact necessary in order to

make the statements made, in the light of the circumstances under which they

were made, not misleading, or (3) to engage in any act, or course of business

which operates or would operate as a fraud or deceit upon any person, in con-

nection with the purchase or sale of any security.” 17 CFR § 240.10b-5 (1951). Per

the misappropriation theory of insider trading, Rule 10b-5 “bars trading on the

basis of information that the wrongdoer converted to his own use in violation of

some fiduciary, contractual, or similar obligation to the owner or rightful posses-

sor of the information.” Barbara Aldave, Misappropriation: A General Theory of Li-

ability for Trading on Nonpublic Information, 13 HOFSTRA L. REV. 101, 122-23 (1984).

. Jonathan R. Macey, From Fairness to Contract: The New Direction of the

Rules Against Insider Trading, 13 HOFSTRA L. REV. 1, 55 (1984) (summarizing the

various motivations behind insider trading jurisprudence and arguing that a

main motivation is the proper allocation of property rights in information).

. United States v. O'Hagan, 521 U.S. 642 (1997) (citing Victor Brudney,

Insiders, Outsiders, and Informational Advantages Under the Federal Securities Laws,

93 HARV. L. REV. 322, 356 (1979)) (additionally discussing why it is not socially

useful to permit those that become in possession of relevant information in the

course of rendering another service to trade on the acquired information ad-

vantages for their personal benefit).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

Collectively, these principles—those prohibiting intermediaries from

using certain information advantages when trading, those prohibiting in-

termediaries from preferentially routing order flow, and those requiring

exchanges to provide traders with fair access to information and speed—

help to protect competition in the electronically traded equities market.

Online advertising, despite also trading on electronic trading venues at

high speeds and facing the same competition issues, benefits from no par-

allel code of conduct.

III. GOOGLE DOMINATES ONLINE ADVERTISING MARKETS BY ENGAGING IN

CONDUCT LAWMAKERS PROHIBIT IN OTHER ELECTRONIC TRADING

MARKETS

Historically, Google was late to enter the programmatic advertising

market, and, when it launched an exchange in the fall of 2009, it faced a

lot of competition.

Right Media’s RMX exchange launched four years

earlier and was processing billions of individual ad spaces daily.

Mi-

crosoft and top Silicon Valley venture funds Sequoia Capital and Draper

. In the late 2000s, the advertising exchange market was competitive. See

generally AdExchanger Staff, supra note 33 (stating that the exchange market “is

one area where Google is playing catch up with Yahoo and their RMX”); Case

No COMP/M.4731 – Google/DoubleClick, Regulation (EC) No 139/2004 Merger

Procedure, Comm’n of the European Communities Decision (Nov. 3, 2008) (dis-

cussing competition in the exchange and network markets between Google, Ya-

hoo!, Microsoft, AOL/Time Warner, and others); Harry Gold, Pros and Cons of Ad

Exchanges, CLICKZ (July 24, 2007), https://perma.cc/M4GZ-JUTF (further discuss-

ing exchange competition from Yahoo!, ContextWeb, AdBrite, and others).

. Michael Arrington, Panama Not Enough to Battle Google: Yahoo Acquires

RightMedia, TECHCRUNCH (Apr. 30, 2007), https://perma.cc/9USU-XFSQ;

AdExchanger Staff, supra note 33 (clarifying that Right Media’s exchange did not

permit real-time bidding on individual impressions at the time). See also Clifford,

supra note 3 (stating that in 2008, Right Media “handle[d] about six billion trans-

actions a day”). See also, Yishay Mansour et al., supra note 4.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

Fisher Jurvetson had their horses in the race as well with three other ex-

changes: AdECN, AdBrite and ADSDAQ, respectively.

Some eight ex-

changes crowded the market, competing against each other on price, auc-

tion mechanics, liquidity and inventory quality.

The same was true

about Google’s entry into the enterprise buying tool market. In 2009,

Google owned Google Ads (then called AdWords), but Google had not

yet launched an enterprise buying tool (DSP), and plenty of firms com-

peted in that market segment.

Despite “playing catch up,” by around 2013, Google’s exchange over-

took the competition to become the largest trading venue for ad space

globally.

Shortly afterwards, Google’s enterprise buying tool called Dis-

. AdExchanger Staff, Ad Exchange AdECN Acquired By Microsoft,

ADEXCHANGER (July 27, 2007), https://perma.cc/9LC7-A6RH; Kristen Nicole,

ContextWeb Launches ADSDAQ for Ad Trading, MASHABLE (May 1, 2007),

https://perma.cc/74VB-PKE5; adBrite, CRUNCHBASE (2020),

https://perma.cc/8G6K-WGGB; Danny Schultz, ContextWeb’s ADSDAQ Exchange

Reaches Record Size and Scope, GOTHAM VENTURES (Nov. 21, 2008),

https://perma.cc/G7VV-DSGT.

. See generally AdExchanger Staff, supra note 33.

. AdExchanger Staff, The AppNexus Reveal: We're A Demand-Side Platform

Says Pres Rubenstein, ADEXCHANGER (Nov. 12, 2009) https://perma.cc/Z92C-53X7

(discussing competition in this market segment from x+1, MediaMath, Invite Me-

dia, CPM Advisors, Adchemy, Fox Audience Network, DataXu, Turn,

AdBuyer.com, and Triggit). A 2011 market research report found the top DSPs

in the market to be those belonging to AppNexus, Turn, MediaMath, and

DataXu. Jason Del Ray, MediaMath in AdAge “MediaMath Declared Winner, but

Who’s the Largest DSP in the Land?,” MEDIAMATH (Dec. 14, 2011),

https://perma.cc/MCX4-ED8S (showing AppNexus was the leading DSP pro-

cessing 26 billion impressions monthly whereas Invite Media processed an esti-

mated 10 billion monthly).

. The first reference to the dominance of Google’s exchange can be found

in the year 2013. John McDermott, Google’s Display Advertising Dominance Raises

Concerns, DIGIDAY (Dec. 2, 2014), https://perma.cc/Z7D3-KRLR (noting that

“Google is the operator of the largest ad exchange, AdX, while also operating a

leading DSP, namely, DBM” (quoting Amir Efrati, FTC Begins Probe of Google’s

Display-Ad Business, WALL ST. J. (May 24, 2013).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

play & Video 360 (“DV360”), an outgrowth of Google’s acquisition of In-

vite Media, also eclipsed the competition to become the most used in the

market.

By 2015, the early exchanges that had initially beat Google to

the punch sold at a discount price or closed their doors.

The market

shares of the DSPs that once led that market segment declined in paral-

lel.

Google, with the language of tech likely obscuring parallels between

ad trading and electronic trading in other markets, rose in the West by

breaching the principles that regulators in the East had crafted over the

years to help safeguard competition in a market with a parallel structure.

A. Google Has Information and Speed Advantages

1. Google Acquires Leading Ad Server DoubleClick

One reason why Google’s exchange and buying tools expediently

surpassed their competitors comes down to an information advantage

they have when buying and selling ads, which traces its origin to the ad

server software company called DoubleClick that Google acquired in

. Google acquired one of the first DSPs, Invite Media, in 2010, which

Google re-launched into the market in 2012 as “DoubleClick Bid Manager” (now

called DV360). Peter Kafka, Google’s Final Price Tag for Invite Media: $81 Million,

ALL THINGS D (June 9, 2010), https://perma.cc/PP5S-AE9J.

. See Zachary Rodgers, Microsoft Retires AdECN, Migrates to AppNexus

RTB Engine, CLICKZ (Mar. 30, 2011), https://perma.cc/9R6A-5ZEL; Anthony Ha,

Ad Exchange adBrite to Shut Down on Feb. 1, TECHCRUNCH (Jan. 28, 2013),

https://perma.cc/2JQD-JRGN; Jeff Fraser, Yahoo Ad Exchange Closes Its Doors to

Publishers, MKTG. MAG (Jan. 13, 2015), https://perma.cc/X3C5-RQQK. The

ADSDAQ exchange became ContextWeb, before PulsePoint acquired it. Today,

PulsePoint appears to be focusing on the healthcare segment and operates header

bidding technology (see infra Part III.B) but does not appear to be operating an

exchange.

. Today, Google’s tool DV360 is the leading one in the market. See gen-

erally CMA FINAL REPORT, supra note 7; AUTORITÉ DE LA CONCURRENCE, supra note

7 (finding that Google’s enterprise buying tool DV360 generates the most reve-

nue and has significant growth).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

2007.

DoubleClick—founded in 1996 and listed on the Nasdaq just two

years later—was the leading ad server globally and claimed about 57%

of the market in the U.S.

A long list of news publishers, including Sports

Illustrated, AOL Online, and The Wall Street Journal, used DoubleClick to

migrate their business from print to digital and to automate the process

of targeting and selling ads.

As an intermediary in this market, DoubleClick was privy to critical

information about publishers’ readers, which publishers use when sell-

ing their ad space on real-time electronic exchanges. One of the most im-

portant pieces of information in this category is the identity of users as-

sociated with publishers’ ad space. To identify publishers’ online readers,

the ad server assigns users a “cookie” to track users by random, unique

identifiers (“user IDs”).

As long as users’ internet browsers permit

. Press Release, Google Inc., Google to Acquire DoubleClick (Apr. 13,

2007), https://perma.cc/JZ58-LLT9.

. Information on DoubleClick publisher ad server market shares in 2008

is readily available. See Abbey Klaassen, Google Leads in Ad-serving Share, ADAGE

(Dec. 18, 2008) (estimating Google’s ad server market share, including Dou-

bleClick and AdSense, totaled 57%); Brockhoff et al., supra note 38 (summarizing

that the European Commission conducted an investigation into the ad server

market and found that DoubleClick was the leading publisher ad server used by

publishers in the European Economic Area, followed by the ad servers belonging

to 24/7 Real Media/OpenAdStream and AdTech/AOL); Statement of Federal

Trade Commission Concerning Google/DoubleClick, No. 071-0170 (F.T.C. Dec.

11, 2007), https://perma.cc/636J-EF2B ( “DoubleClick is the leading firm in the

third party ad serving markets. It faces competition from several significant com-

petitors in the United States, such as 24/7 Real Media, aQuantive, and Value-

Click.”). Additionally, many smaller publishers also did (and still do) use

Google’s AdSense product to perform ad serving functions rather than licensing

a standalone ad server such as Google’s DoubleClick product.

. See What Customers are Saying About DART for Publishers, DOUBLECLICK

(Sept. 1, 2006), https://perma.cc/6FW9-Z2LB; Louise Story & Miguel Helft, Google

Buys DoubleClick for $3.1 Billion, N.Y. TIMES (Apr. 14, 2007),

https://perma.cc/UCJ7-X49Y.

. For a more thorough explanation of how cookies work, see AdSense

Help, DoubleClick Cookies, GOOGLE (Mar. 1, 2017),https://perma.cc/J5MT-MGFD .

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

tracking through cookies by default, these IDs become a proxy for users’

identity online.

Because DoubleClick was the number-one ad server in

the market, it had cookied the most online users and was the leader in

online identity information.

Tracking users with identifiers is of high value to advertisers who

buy ad space for myriad reasons. At a basic level, user IDs

are used to

recognize users, remember information about their prior visits, and more

effectively target ads.

For example, if John visits The New York Times on

Monday and reads the article “How to Do Laundry,” The Times’ ad server

can assign John an ID, such as 1Q2W3E.

When John returns on Friday,

the ad server recognizes John as user 1Q2W3E who read about laundry

earlier in the week. The ad server can use that information to show John

. The way that user tracking works for mobile applications is different

than the way it works for advertising on desktop or mobile web. For ads that

display inside of mobile applications, the mobile device operating system (such

as Apple or Android) assigns a user’s device an advertising ID, and this ID is

signaled to exchanges along with information about specific ads for sale. See

MOBILE IDENTITY WORKING GROUP, MOBILE IDENTITY GUIDE FOR MARKETERS: A

BEST PRACTICES PRIMER FOR MOBILE & CROSS-DEVICE MARKETING (IAB ed., 2017);

Authorized Buyers Help, Target Mobile Apps with IDFA or AAID, GOOGLE,

https://perma.cc/6CK6-D63K (“[B]id requests from Android devices pass the

AAID [Android Advertising ID], . . . [a] device-specific, unique, resettable ID for

advertising . . . .”); Google AdMob Advantage, Discover the AdMob Advantage,

GOOGLE, https://perma.cc/7M4S-V233 (last visited Nov. 16, 2020) (explaining

how the mobile app advertising vendor AdMob shares users’ identities with ex-

changes).

. We can infer that DoubleClick had IDs for the most users online be-

cause DoubleClick had partnerships with the most websites permitting Dou-

bleClick to track websites’ users using cookies. See Balachander Krishnamurthy

& Craig Wills, Privacy Diffusion on the Web: A Longitudinal Perspective, INT’L

WORLD WIDE WEB CONF. COMM. (Apr. 2009), https://perma.cc/SGS3-Y5TM (study

of 1200 sites showing doubleclick.net as the number one tracker between 2005-

2008).

. STEVE JONES, ENCYCLOPEDIA OF NEW MEDIA: AN ESSENTIAL REFERENCE

TO COMMUNICATION AND TECHNOLOGY 95-96 (2003).

. Jolie Kerr, How to Do Laundry, N.Y. TIMES, https://perma.cc/76PH-

CVH8 (last visited Nov. 16, 2020).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

an ad for detergent even though he might be reading an article about the

weather. Access to the ID also permits advertisers bidding through ex-

changes to access information they already have about John or query

more information about him from a third-party data vendor.

While

John’s page is still loading, it might query its internal customer records,

or connect with a provider of audience data, to learn that John has sensi-

tive skin and to refine its decisioning and targeting algorithms to show

the ad for unscented detergent instead.

At the time of the Google-DoubleClick merger, the Federal Trade

Commission (FTC) and others had faith that Google would not leverage

its control of publishers’ primary ad server to distort competition in the

electronic trading market. The FTC rejected prescient concerns about data

and competition raised by public interest groups and FTC Commissioner

Pamela Jones Harbour (dissented), stating that the evidence showed that

“it was unlikely that Google could manipulate DoubleClick’s third-party

ad serving products in a way that would competitively disadvantage

Google’s competitors in the ad intermediation market.”

In covering the

. Advertisers commonly query data from data management platforms

(DMPs), which are specialized software companies that warehouse, analyze, and

process data. For more on this, see Maciej Zawadziński, What is a Data Manage-

ment Platform (DMP) and How Does It Work?, LINKEDIN (Sept. 1, 2015),

https://perma.cc/W7J8-M49T.

. Knowing the IDs of users is also used for non-targeting purposes. For

example, advertisers do not want to inundate users with the same ad, and know-

ing the user’s ID allows them to limit how many times they show the same ad to

the same user. This feature of online advertising called “frequency capping” has

been around since the early 2000s. Additionally, knowing the IDs of users view-

ing ads allows advertisers to measure the effectiveness of a campaign, since they

allow advertisers to track whether a user took a subsequent action, such as click-

ing on an ad, signing up for a service, or purchasing a product. JONES, supra note

76.

. Complaint and Request for Injunction, Request for Investigation and

for Other Relief at 9-10, In the Matter of Google, Inc. and DoubleClick, Inc.,

(F.T.C. April 20, 2007) (urging the FTC to open an investigation into Google’s

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

merger for the wider public, The New York Times was equally optimistic.

It acknowledged that Google’s DoubleClick division would have con-

flicts of interest with Google’s “Nasdaq-like exchange,” but suggested

that publishers and advertisers would “jump ship” if Google failed to

properly manage its conflicts.

Google’s representations to government bodies furthered the belief

that Google would properly manage its conflicts of interest, including

when it came to data. To Congress, Google’s general counsel assured that

DoubleClick “data is owned by the customers, publishers and advertis-

ers, and DoubleClick or Google cannot do anything with it.”

Indeed,

proposed acquisition of DoubleClick and to halt the merger absent assurances

that Google will not merge DoubleClick datasets with Google datasets and en-

gage in other data conduct that could hurt consumers); Second Filing of Supple-

mental Materials in Support of Pending Complaint and Request for Injunction,

Request for Investigation and for Other Relief at 4, In the Matter of Google, Inc.

and DoubleClick, Inc., (F.T.C. Sept. 17, 2007) (asking the FTC to condition the

merger on Google and DoubleClick “maintaining separate databases of user in-

formation”); Pamela Jones Harbour, Dissenting Statement in Google/Dou-

bleClick F.T.C. File No. 071-0170 (F.T.C. 2007), https://perma.cc/Q57H-3YT8 (spe-

cifically warning that if Google and DoubleClick are permitted to merge without

conditions, the firm may merge data between Google and DoubleClick, to nega-

tively harm both consumer privacy and competition; also stating that the merger

“has the potential to profoundly alter the 21st century Internet-based economy –

in ways we can imagine, and in ways we cannot”); Federal Trade Commission

Closes Google/DoubleClick Investigation, FED. TRADE COMM’N (Dec. 20, 2007),

https://perma.cc/B5JD-QVNC [hereafter FTC Closes Investigation] (additionally

stating that the FTC would “closely watch these markets and, should Google en-

gage in unlawful tying or other anticompetitive conduct, the Commission in-

tends to act quickly”).

. Story & Helft, supra note 72(“The sale also raises questions about how

Google will manage its existing business and that of the new DoubleClick unit

while avoiding conflicts of interest. If DoubleClick’s existing clients start to feel

that Google is using DoubleClick’s relationships to further its own ad network,

some Web publishers or advertisers might jump ship.”).

. An Examination of the Google-DoubleClick Merger and the Online Advertis-

ing Industry: What Are the Risks for Competition and Privacy?: Hearing Before the Sub-

committee on Antitrust, Competition Policy and Consumer Rights, 110th Cong. (2007),

https://perma.cc/39UD-MKDZ [hereinafter The Google-DoubleClick Merger].

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

with ownership of such data contractually vested in the buyers and

sellers in this market, to do anything with that data would be, by defini-

tion, a misappropriation.

However, the FTC’s confidence was misplaced. DoubleClick and

Google were already under fire for how they handled information and,

in an electronically traded equities market, regulators must proactively

manage conflicts of interest in order to ensure a company will not misuse

inside information. When financial intermediaries are allowed to have

conflicts, ethical walls and fiduciary (or best interest) duties restrict them

from using their customers’ information for their own financial gain.

There was also a problem with jumping ship. As inventory management

software integrated into back-end billing systems, it was a jump that

some compare to changing engines mid-flight.

Thus, in a 4-1 vote, the FTC approved the merger without conditions.

However, the concerns raised by the public interest groups and FTC

Commissioner Harbour would come to fruition. Google’s conflicts of in-

terest would get the better of it. When it officially launched its exchange

in 2009, Google began to restrict publishers’ and advertisers’ ability to

access their DoubleClick data and to reserve for its own trading divisions

an “essential” information advantage.

. Ratko Vidakovic, The Beginner’s Guide to Header Bidding, ADPROFS (Mar.

21, 2019), https://perma.cc/98WQ-G42C (“As a publisher, replacing your primary

ad server is not a trivial task. Think of it like doing a mid-flight engine swap on

an airplane. Except that it’s your revenue engine. It’s hard to imagine many pub-

lishers wanting to take such a risk.”); Tim Cross, Why Don’t More Publishers Switch

Ad Servers? VIDEO AD NEWS (June 6, 2019), https://perma.cc/9MD7-GU27.

. Adsense Help states the following: “Cookie ID in each DoubleClick

cookie is essential to these applications. For example, DoubleClick uses cookie

IDs to keep a log of which ads show to which browsers. When it’s time to serve

an ad to a browser, DoubleClick can use the browser’s cookie ID to check which

DoubleClick ads have already been delivered to that particular browser. That’s

how DoubleClick avoids showing ads the user has already seen. In the same way,

cookie IDs allow DoubleClick to log conversions related to ad requests—such as

when a user views a DoubleClick ad and later uses the same browser to visit the

advertiser’s website and make a purchase.” AdSense Help, supra note 73.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

2. DoubleClick Ad Server Starts to Play Favorites When Sharing

User Identity

Competition in the exchange market, one step removed from the ad

server that routes to exchanges, depends on the ability to recognize users

associated with ad space for sale.

Without access to user IDs, the de-

mand for ad space wanes and their clearing prices on exchanges drop

dramatically as a result. Recent pricing and bid data from various ex-

changes illustrate the point. For example, a 2018 Google study reports

that the prices for ad space trading on Google’s exchange drop by 50+%

when advertisers cannot identify users associated with ad space for sale.

Relatedly, Index Exchange has shared that the number of bids for ad

space on Mozilla Firefox pages dropped by 38% after the internet browser

started blocking cookies.

That is, without knowing the identity of the

user, advertisers often sit out of ad auctions altogether. Although some

. See generally AdSense Help, supra note 73; Ronan Shields, Inside the IAB

Tech Lab’s New Consumer Privacy Proposals, ADWEEK (Sept. 19, 2019),

https://perma.cc/3PP2-LS47 (discussing how access to user IDs inside bid re-

quests is what permits firms to conduct basic programmatic advertising pro-

cesses).

. Deepak Ravichandran & Nitish Korula, Effect of Disabling Third-Party

Cookies on Publisher Revenue, GOOGLE (Aug. 27, 2019), https://perma.cc/79K4-

RLHJ (“We disabled access to cookies for a small fraction of randomly selected

users (the treatment group). We observed that for the top 500 global publishers,

average revenue in the treatment group decreased by 52%, with a median per-

publisher decline of 64%.”). See also Garrett Johnson et al., Consumer Privacy

Choice in Online Advertising: Who Opts Out and at What Cost to Industry? (Simon

Business School, Working Paper No. FR 17-19, 2019) (“We estimate that opt-out

consumers fetch 52% lower prices on the exchange than the counterfactual prices

they would fetch with tracking cookies.”).

. Mike O'Sullivan, The Impact of Mozilla Firefox: The Change Is Now, INDEX

EXCHANGE (Oct. 30, 2019), https://perma.cc/W5J3-GT9V; Ross Benes, Publishers

Are Already Feeling Pain from Apple’s Move Against Ad Tracking, DIGIDAY (Oct. 15,

2017), https://perma.cc/4XAZ-S8NE (reporting one publisher receives 15% less

bids for Safari impressions).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

advertisers do bid on this type of cookie-less ad space, data from the Ru-

bicon exchange shows that the clearing prices on Safari pages drop by

60% overall.

To maximize competition for ad space in auctions, and in turn their

clearing prices, all parties bidding on inventory need to recognize users

associated with the ad space for sale. Publishers could facilitate this out-

come by sharing their ad server user IDs with all bidders in a non-dis-

criminatory manner. After all, ownership of DoubleClick data rested in

publishers and advertisers, and Google could not do anything with it.

But ownership rights in the ad trading market are malleable: they

vest from negotiating power and contract rights. When Google launched

an exchange into a crowded market in 2009 and its ad server developed

a conflict, the ad server started to restrict publishers’ and advertisers’

ability to access the ad server user IDs.

In fact, Google restricted how

any party other than Google could access them. It did this by hashing the

. Tom Dotan, Apple’s Ad-Targeting Crackdown Shakes Up Ad Market,

INFORMATION (Dec. 9, 2019), https://perma.cc/69YT-QU38 (Rubicon Project CEO

Michael Barrett explaining that “[t]he allure of a Safari user in an auction has

plummeted” and “[t]here’s no easy ability to ID a user”). See also Yoel Minkoff,

Safari Privacy Features Disrupt Ad Market, SEEKING ALPHA (Dec. 10, 2019),

https://perma.cc/AYQ9-HJ6N.

. See AdExchanger Staff, AdExchanger.com Q&A with Google DoubleClick

Ad Exchange's Mohan and Spencer [hereinafter AdExchanger Q&A], ADEXCHANGER

(Sept. 21, 2009), https://perma.cc/J3WN-YGPV. Before Google’s acquisition of

DoubleClick, DoubleClick faced no conflict of interest, which may explain why

DoubleClick’s policy towards sharing user IDs was more neutral back then.

Note, prior to 2003, DoubleClick had a division that licensed ad management

software and another that acted as a sales broker for publishers who were part

of the “DoubleClick Network.” However, this changed in 2002 when Dou-

bleClick divested its ad network division to focus on licensing ad serving soft-

ware. See DoubleClick Inc., Registration Statement (Form S-1) (Nov. 18, 1998),

https://perma.cc/P3M7-322W (discussing both lines of business activity at the

time of the DoubleClick IPO); DoubleClick Inc., Annual Report (Form 10-K),

https://perma.cc/ZX55-R9R9 (discussing divestiture in 2002 through a series of

transactions).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

user IDs, which is a process of scrambling characters based on a mathe-

matical formula.

If John’s ID was 1Q2W3E, Google would share a dif-

ferent ID with the Iowa news publisher The Register (e.g., 0P9O8I) than it

would share with the advertiser (e.g., MM99NN), thereby preventing

both entities from realizing that the person in question was in fact the

same user.

The change did not go unnoticed. Industry publication AdExchanger

interviewed senior Google product manager Scott Spencer about the

change, asking whether the new approach “break[s] the universal cookie

model that helps create efficiencies for advertisers and, ultimately, more

frictionless advertising for the user . . . .”

Without delving into specifics,

Spencer replied that a concern for user privacy motivated the change.

At the same time that Google’s ad server started to restrict access to

these IDs for privacy reasons, Google permitted its own exchange and

buying tools to access them by default. First it was Google’s buying tool

for small businesses—Google Ads, which was then called AdWords.

. Specifically, they are hashed differently for each publisher, advertiser,

and intermediary. See AdExchanger Q&A, supra note 89 (discussing Google’s de-

cision to start hashing IDs); Google Developers, Cookie Matching, GOOGLE (Jan. 5,

2014), https://perma.cc/A7WF-SAL7 ( “For buyers, Google identifies users using

a buyer-specific Google User ID which is an encrypted version of the dou-

bleclick.net cookie, derived from but not equal to that cookie.”); Google Ad Man-

ager Help, Data Transfer Fields, GOOGLE (2020), https://perma.cc/U62K-V5EX

(“The DoubleClick cookie ID associated with the user, encrypted.”). One needs a

“key” to unscramble the user id, but Google does not provide keys for that pur-

pose. Campaign Manager Help, User ID Encryption, GOOGLE (2020),

https://perma.cc/GCE2-TGWE (“You will never be able to decrypt user IDs, and

Google will not disclose the encryption method. No encryption keys will ever be

provided to any Campaign Manager customer or any third-party partner.”).

. AdExchanger Q&A, supra note 89.

. Note, the buying tool is called AdWords, but the product for publish-

ers is called AdSense. Privacy Policy, GOOGLE (Mar. 11, 2009),

https://perma.cc/VQ8C-ZB79 ( “Google uses the DoubleClick advertising cookie

on AdSense partner sites . . . .”); AdSense Help, AdSense Program Policies, GOOGLE

(July 8, 2016), https://perma.cc/9FGA-3S4Q (“Google uses the DoubleClick

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

Then it was Google’s exchange. When Google launched an exchange in

the fall of 2009, the exchange recognized website visitors by their Dou-

bleClick IDs.

Finally, it was Google’s DSP, the enterprise buying tool.

When Google launched its DSP, DoubleClick Bid Manager (now called

DV360), in 2012, this software recognized users by their DoubleClick IDs

as well.

The change to hashing user IDs for everyone other than Google re-

served an information advantage for Google when it came to selling web-

sites’ ad inventory. Specifically, Google’s exchange benefits from better

recognizing online users, which allows it to better facilitate trades be-

tween websites and buyers, compared to rival exchanges.

An advertiser

cookie on publisher websites displaying AdSense for content ads.”); News from

Google, Google Expands Advertising Monetization Program for Websites, GOOGLE

(June 18, 2003), https://perma.cc/YR2Y-HYFL (announcing and explaining the

AdSense product).

. AdExchanger Q&A, supra note 89.

. Payam Shodjai, Getting Real with DoubleClick Bid Manager,

DOUBLECLICK ADVERTISER BLOG (Oct. 24, 2012), https://perma.cc/S9N6-TZ3W (an-

nouncing the launch of Google’s DSP DoubleClick Bid Manager and the sunset-

ting of Invite Media and that user id lists in advertisers’ ad server are available

in DoubleClick Bid Manager “without any data loss, leading to better reach and

performance for remarketing campaigns”); Jason Miller, Constantly Innovating:

the DoubleClick Digital Marketing Platform, DOUBLECLICK ADVERTISER BLOG (Sept.

19, 2012), https://perma.cc/N8L6-NTRE (explaining that buyers can easily target

the list of users that saw their Google Search ads through Google’s DSP since the

integration between the two products does not require advertisers to install new

site tags); Google Developers, Cookie Matching, GOOGLE , https://perma.cc/WL6Q-

GYPD (last visited Nov. 16, 2020) (“For itself, Google identifies users with cook-

ies that belong to the doubleclick.net domain under which Google serves ads.”).

. For example, suppose John was assigned a DoubleClick ID of 1Q2W3E

but had not been assigned a Right Media RMX or Google Ad Exchange ID. If

neither the Right Media nor the Google exchange could recognize John by an ID,

each might have sold an initial ad targeted to John for a low $1 CPM price, which

stands for cost per one thousand impressions and is one common advertising

pricing model. After the acquisition, Google’s exchange could access John’s Dou-

bleClick ID by default, which allowed it to sell the initial same ad targeted to

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

that is a client of the DoubleClick ad server has to go through Google’s

exchange in order to efficiently buy more ads targeted to the same users

because only Google tracks users by the same ID.

On the ground, this

worked to lock-in clients of the DoubleClick ad server to Google’s ex-

change and buying tools, fueling Google’s reputation in the industry as a

“walled garden.”

3. Google-Owned Intermediaries Have an Information Advantage on

Google’s Exchange

Google’s asymmetric approach to sharing websites’ DoubleClick user

IDs also distorts competition between the buying tools competing to pur-

chase ad space from Google’s exchange (e.g., DV360 and Google Ads ver-

sus rival buying tools). This helps explain how Google eventually lever-

aged its dominance in the ad server market, then the exchange market,

into dominance in the buying tool market too, to the detriment of adver-

tisers. Google’s exchange shares users’ DoubleClick IDs with the Google-

John for a higher price than could Right Media’s exchange RMX. Google’s ex-

change benefits from piggybacking off the Google DoubleClick ad server IDs in

other ways too. See infra note 103.

. The ability to share IDs facilitates better interoperability between

Google and non-Google tools. For example, an advertiser using the DoubleClick

ad server might have shared a list of user IDs directly with a publisher or non-

Google intermediary to target more ads to the same users. This old advertising

contract between DoubleClick and Compaq Computer from 1999 corroborates

the understanding that before Google’s acquisition of DoubleClick, DoubleClick

permitted its customers more discretion over how to use their DoubleClick ad

serving software data. DoubleClick Inc. and Compaq Computer Corp., supra note

38 (stating that “Compaq has the sole and exclusive right to use all data derived

by its use of the DART [ad serving software] Service, for any purpose related to

Compaq's business with Advertisers”).

. See generally John McDermott, Welcome to the ‘Walled Garden’ Era of Ad

Tech, DIGIDAY (Feb. 12, 2015), https://perma.cc/8N99-WRZ3.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

owned buying tools.

However, when sending bid requests to non-

Google intermediaries, Google’s exchange shares a different ID value.

The information asymmetry, flowing from Google’s number-one ex-

change to the buying tools that have a seat to bid, distorts competition

between bidders. To identify users associated with ad space for sale on

Google’s exchange, the non-Google intermediaries have to use cookies to

assign users a new set of proprietary IDs and must synchronize their IDs

with Google’s hashed ones through a process called cookie syncing.

100

The problem is that the cookie syncing process is inherently inefficient.

101

In addition to increasing page latency for users, the new IDs belonging to

the non-Google intermediaries will not always match with Google’s

hashed IDs, resulting in advertisers sometimes not knowing the identity

of users associated with ad space trading on Google’s exchange. Industry

. Google Developers, supra note 90.

. AdExchanger Staff, supra note 90; Google Developers, supra note 90;

Google Ad Manager Help, supra note 90; Campaign Manager Help, supra note

90.

100

. To give a simple example, if Google’s exchange sends a bid request

indicating there is an ad available for user 0P9O8I, the non-Google buying tool

will initially not know who that is because it tracks the user by a different ID

(e.g., BB88VV). To solve this problem, Google’s exchange and the non-Google

buying tool swap their proprietary IDs for the same user through “cookie sync-

ing.” After a cookie sync has occurred successfully, the trader will know bid re-

quests for user 0P9O8I correspond to user BB88VV in its own records. See Google

Developers, supra note 90; Ronan Shields, The Trade Desk Rolls Out Its Unified ID

to Take on the Walled Gardens, ADWEEK (Oct. 24, 2018), https://perma.cc/C6XW-

DAYR; Maciej Zawadziński, What Is a Cookie Syncing and How Does it Work?,

CLEARCODE BLOG (Dec. 14, 2015), https://perma.cc/AWA4-BRK8; Infrastructure

Options for RTB Bidders (Part 4), GOOGLE CLOUD , https://perma.cc/LHE5-C97E

(last visited Nov. 16, 2020) (explaining cookie syncing and that syncing can also

happen outside of bid requests).

101

. Cookie syncing also has negative consumer welfare effects, including

consumer privacy implications and slower page loads. See Shields, supra note

100.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

100

insiders call these “blind” spots

102

and quantify the attrition as “loss

rates.”

103

When blind, advertisers bid less or not at all.

102

. See generally SSP to DSP Cookie Syncing Explained, AD OPS INSIDER

(May 1, 2011), https://perma.cc/7KY2-QZEQ; Ben Kneen, What Exactly Is "Data

Loss" and "Cookie Loss" in AdTech?, QUORA (June 22, 2015),

https://perma.cc/P4NY-7AA7 (explaining loss rates); Chris O’Hara, The Match

Game: Match Rates Are the New Click-Through Rates, ADEXCHANGER (Nov. 30,

2015), https://perma.cc/T3WB-879X (explaining that match rates of 60% are

“overall pretty good” and of 70% are “excellent”).

103

. Loss rates can occur because of technical glitches during the synchro-

nization process, because users block or delete cookies, or because companies

may not have enough time during page loads to sync with other companies in-

volved in the trade. For instance, when users delete their cookies and the associ-

ated user IDs assigned to them, exchanges and buying tools race to assign those

users new cookies and new IDs. Since so many consumers use Google Search and

Gmail, and so many websites use Google’s ad server product, the company that

can do this fastest is often Google. Because Google is the fastest, this can mean

that Google’s exchange and buying tools can re-identify users the fastest too. On

the other hand, competitors have to re-assign users new cookies and synchronize

new IDs with Google’s, resulting in higher loss rates. The Most Popular Email Pro-

viders in the U.S.A., SHUTTLECLOUD (Aug. 11, 2016), https://perma.cc/HET2-

DGGK (stating that Google’s Gmail service has 130+ million U.S. users and has

53% share of the U.S. email market). Moreover, Google’s buying tools may also

have a leg-up when it comes to user identity because of Google’s ability to avoid

third-party cookie blocking (by ad blockers and internet browsers). Some ad

blockers and browsers do not block Google Analytics first party cookies, ena-

bling Google to get around the blocking by sharing data between the non-

blocked Google Analytics cookies and its advertising divisions. Which Ad Blockers

Are Blocking Google Analytics?, ANALYTICAL 42 (Aug. 10, 2016),

https://perma.cc/YVV2-D2MF (discussing how ad blockers do not block Google

Analytics cookies); Google Developers, Google Analytics Cookie Usage on Websites,

GOOGLE ANALYTICS, https://perma.cc/JGT4-5SDU (last visited Sept. 28, 2020);

Google Support, About Remarketing Audiences in Analytics, GOOGLE,

https://perma.cc/4MR3-CESG (last visited Sept. 28, 2020) (both discussing how

Google shares data from Google Analytics with Google advertising divisions);

Parker Robben, Here Are Your Options for Google Ads Tracking with Apple’s ITP 2.0,

METRICTHEORY (Sept. 18, 2018), https://perma.cc/LGX4-YE9M (explaining the

linking of data pools between first-party cookies from Google Analytics and

Google Ads and Google Tag Manager).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

101

The Google-owned buying tools do not suffer from this information

asymmetry. DV360 (enterprise) and Google Ads (small business) more

frequently recognize the identity of users associated with ad space on

Google’s exchange, which means they more frequently make informed

decisions about the value of inventory for sale and return informed bids

on behalf of advertisers. This information advantage, reserved by Google

for Google, is something that Google markets to advertisers and publish-

ers as a synergy for using Google all the way through the trade: Google’s

intermediary on the sell-side (i.e., ad server), Google’s exchange, and

Google’s intermediary on the buy-side (i.e., the buying tools).

104

. See Google Ad Manager Help, supra note 46 (explaining that advertis-

ers “minimize” losses related to “cookie matching” and “there is a higher likeli-

hood they’ll find impressions that meet their targeting criteria” when they use

DV360 or Google Ads); Shields, supra note 100 (explaining generally that adver-

tisers spend disproportionately with Google because it is comparatively easier to

target audiences). Note, Google also explains to publishers that they should ben-

efit from greater “auction pressure” because of the information advantages be-

tween Google’s exchange and Google’s buying tools. Google Ad Manager Help,

supra note 46. However, the expected higher bid prices for publishers may be

tempered by new “Unified Pricing Rules” that Google introduced in April 2019.

These rules now prohibit publishers from routing their ads through Google’s ex-

change at prices higher than they route them through competing exchanges. For

further discussion on Google price parity restrictions and their effect on compe-

tition see infra Part III.B and Groth, infra note 183. Consequently, although

Google states the information advantages of Google buying tools should garner

higher bids for publishers, Google's Unified Price Rules (which are a condition

of using Google's ad server) can make sure that ads are sold at the lowest price

across the market. Jason Bigler, An Update on First Price Auctions for Google Ad

Manager, GOOGLE BLOG (May 10, 2019), https://perma.cc/F8JG-WVNA (stating

that publisher price floor will hereinafter “be applied to all partners equally, and

cannot be set for individual buying platforms”); Google Ad Manager Help, Uni-

fied Pricing Rules, GOOGLE SUPPORT (Sept. 28, 2020) https://perma.cc/A2YS-7WN6

(discussing beta rules); Jessica Davies, “It’s a Shakedown”: Everything You Need to

Know about Google’s “Unified Pricing” Product Changes, DIGIDAY (Apr. 25, 2019),

https://perma.cc/U6A2-C6VU (stating that publishers had historically set higher

floor prices for Google’s exchange); Sarah Sluis, Publishers Lash Out Against Google

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

102

4. Consumer Privacy Offered as Reason for Information Asymmetry

Google’s excuse for sharing information in an asymmetric fashion is

consumer “privacy,” but its reasoning does not hold water. Supposedly,

Google does not want firms to combine consumer data records to-

gether.

105

By sharing John’s data with Company A as belonging to user

1Q2W3E and sharing John’s data with Company B as belonging to user

0P9O8I, Company A and B cannot join their respective information about

John together, to create a more comprehensive profile of his internet ac-

tivity. However, this is precisely what Google does. By migrating its ex-

change and buying tools to the same DoubleClick cookie, and by merging

consumer data sets between Google business divisions, Google creates

deeper and deeper profiles of consumers’ internet activity.

Furthermore, Google contract terms already prohibit companies from

combining their Google data records.

106

As a result, scrambling IDs has

over “Unified Pricing” Changes, ADEXCHANGER (Apr. 18, 2019),

https://perma.cc/Z3Z7-AA3Z (discussing reception of Google’s Unified Pricing

changes).

105

. In 2009, Google did not explain what it meant by consumer privacy.

However, online advertising industry executives have shared that Google has

stated concern about companies combining consumer data records. Google

shares this reasoning in a recent submission to the U.K. competition commission.

See AdExchanger Q&A, supra note 89; CMA FINAL REPORT, supra note 7 at 409 (dis-

cussing how Google does not share granular consumer information because

“[i]mposing consistent transaction IDs raises potential privacy concerns” by al-

lowing advertisers to join user data). Further challenging Google’s rationale is

the fact that companies can obtain user consent for this practice and the fact that

companies can combine these types of data records together after-the-fact

through various database matching and merging techniques. Google has also

raised concern about companies possibly using consistent user IDs to merge us-

ers’ internet activity with users’ real identity. However, Google itself engages in

precisely this practice. See CMA FINAL REPORT, supra note 7 at 409 (discussing

how Google states consistent IDs could permit “individual users to be identi-

fied”); Julia Angwin, Google Has Quietly Dropped Ban on Personally Identifiable Web

Tracking, PROPUBLICA (Oct. 21, 2016), https://perma.cc/QN9L-2MHB.

106

. Authorized Buyers Program Guidelines, GOOGLE,

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

103

the added effect of interfering with competition. An advertiser that uses

Google’s DoubleClick ad server now has a harder time using a non-

Google buying tool because the two tools operate on different user IDs.

The privacy imperative has turned out to be Google’s Trojan Horse

in extracting synergies from DoubleClick. Since 2009, Google has contin-

ued to restrict competitors’ access to trading data in various ways for pri-

vacy reasons, without imposing a parallel limitation on Google’s use of

the same information.

107

For example, in 2018, Google stopped allowing

https://perma.cc/BL9Y-3Z9H (last visited Sept. 28, 2020) (stating that buyers that

do not win a specific impression are prohibited from using data associated with

that impression to create user lists, profile users, or associating said data with

third-party data; also prohibiting buyers that do win specific impressions from

retaining impression data outside of Location ata for any use other than for fore-

casting inventory through Google’s exchange). Additionally, other exchanges do

not necessarily approach privacy in this way. See, e.g., AdExchanger Staff, supra

note 90 (comparing Right Media exchange’s practice of storing advertisers ID in

the RMX cookie with Google’s exchange practice of sharing unique hashed user

IDs); Outgoing Bid Request to Bidders, XANDR, https://perma.cc/5VWW-7B9L (last

visited Sept. 28, 2020) (discussing how the Xandr exchange shares a unique 64-

bit ID for a specific user if the seller permits the sharing of user IDs and that IDs

remain stable until the user clears his or her cookies); List of Supported OpenRTB

Bid Request Fields, INDEX EXCH., https://perma.cc/KT5F-PWAD (last visited Nov.

17, 2020) (stating the Index Exchange IX shares with buying tools a “IX static

identifier for this user”); David Simerly & Catherine Racette, OpenRTB 2.3 Bid

Request Objects and Parameters, PUBMATIC (Oct. 26, 2020), https://perma.cc/HSU3-

PHH7 (“Exchange-specific ID for the user. At least one of ID or buyeruid is rec-

ommended.”).

107

. See, e.g., James Hercher, How We Got Here: A Look Back at the Privacy

Changes That Reshaped Google, ADEXCHANGER (Oct. 7, 2019 12:36 AM),

https://perma.cc/MJQ6-V4ES; Eric Berry, How Will Google’s Move to Restrict Third-

Party Cookies Affect Publishers?, ADEXCHANGER (Apr. 5, 2019),

https://perma.cc/MMG8-Z9QN. Conversely, Google has decreased consumers’

privacy with respect to Google itself over the years. For example, Google today

owns and operates the most widely distributed online trackers. These trackers

are also particularly invasive, due to a privacy change Google made in 2016 that

permits Google to combine data obtained from its DoubleClick trackers with in-

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

104

advertisers to access the encrypted user IDs from campaign reports,

which advertisers need to access in order to hire non-Google campaign

measurement firms.

108

Google stated that the change was made to protect

user privacy and alluded to Europe’s new General Data Protection Reg-

ulation (GDPR) privacy legislation, though no provision of GDPR ap-

pears to require Google to take this approach.

109

Meanwhile, advertisers,

provided they stay within Google’s Walled Garden and use Google’s new

Ads Data Hub product, can still access these IDs.

110

formation it knows about consumers’ real identity, in order to conduct non-anon-

ymized surveillance of consumers. Steven Englehardt & Arvind Narayanan,

Online Tracking: A 1-Million-Site Measurement and Analysis, PRINCETON WEB

CENSUS, https://perma.cc/Y2LE-QQR6 (last visited Sept. 28, 2020) (empirical

study of 1 million sites showing that the top five third-party domains distributed

on websites online and twelve of the top twenty are owned by Google; stating

the top five are google-analytics.com, gstatic.com, doubleclick.net, google.com,

and fonts.googleapis.com); Angwin, supra note 105.

108

. Alison Weissbrot, 4 Ways the Death of the Cookie in Chrome Could Affect

Marketers, ADEXCHANGER (Jan. 17, 2020), https://perma.cc/NWM7-DB8F.

109

. See also Understanding the Digital Advertising Ecosystem and the Impact

of Data Privacy and Competition Policy: Hearing Before the S. Comm. on the Judiciary,

116th Cong. (2019) (questions for the record from Senator Lindsey O. Graham to

Mr. Brian O’Kelley) (Brian O’Kelley, former CTO of the first advertising ex-

change at Right Media and later founder of AppNexus, stating: “In effect, GDPR

[EU privacy regulation] gave Google to take anti-competitive actions on the basis

of privacy that made it almost impossible for competitors to operate”).

110

. Google still shares the IDs with Ads Data Hub, a product of Google

Cloud, which means advertisers need to use Data Hub to run these calculations.

Weissbrot, supra note 108; Seb Joseph, Advertisers See Google’s New DoubleClick ID

Rules Cementing Its Dominance, DIGIDAY (Oct. 5, 2018), https://perma.cc/C5KX-

WP28; George Slefo, Google’s Removal of DoubleClick ID Presents Litany of Issues for

Brands, Agencies, ADAGE (May 8, 2018), https://perma.cc/QA7P-7CND. “Some

have noted that Google’s Ads Data Hub, Google’s DMP-like offering, will retain

use of the DID – presenting it with a nearly incomparable advantage. Some agen-

cies have already gone on the record questioning this change and what it means

for their clients – noting that it’s nearly impossible for a marketer to live in a

world completely independent from the Google stack given their dominance in

the DSP, ad server and site analytics world and questioning what this will mean

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

105

Further complicating matters, in January of 2020, Google’s internet

browser Chrome (61% market share) announced that it would block the

technology of third-party cookies altogether within two years.

111

Com-

peting exchanges and buying tools will soon no longer be able to use

cookies to assign users IDs for the purpose of buying and selling ads.

112

It is not clear how the market will adjust to these changes. At the same

time, there is concern that the changes will disproportionately impact ri-

vals because Google has alternative ways of deducing users’ identity.

113

for choice in the marketplace.” Sonja Kristiansen, Ending DoubleClick's ID Export,

MEDIUM: LIFT LETTERS (May 6, 2018), https://perma.cc/646P-DC7G. Additionally,

in 2019, Google removed descriptive information about the pages that ad space

will appear on (e.g., “health”) from bid requests under the guise of privacy con-

cerns. Here, Google’s stated concern made little sense to people in the industry

because Google’s exchange still shares the user’s page URL (e.g.,

webmd.com/hiv-aids), which is much more intrusive on a user’s privacy. Chetna

Bindra, Additional Steps to Further Safeguard User Privacy, GOOGLE BLOG (Nov. 14,

2019), https://perma.cc/SZ4N-Y3G7; Lara O’Reilly, Ad Tech Industry Questions In-

tentions Behind Google’s Latest Privacy Moves, DIGIDAY (Nov. 19, 2019),

https://perma.cc/M8S9-8AZQ.

111

. Justin Schuh, Building a More Private Web: A Path Towards Making Third

Party Cookies Obsolete, CHROMIUM BLOG (Jan. 14, 2020), https://perma.cc/TR97-

8L34 (announcing Chrome changes); Desktop Browser Market Share in United States

of America, STATCOUNTER (Jan. 2020), https://perma.cc/6QTM-KNAN (estimating

Google has 61% market share in the U.S.).

112

. See Berry, supra note 107; John McCarthy & Andrew Blustein, As

Google Chrome Crumbles the Third-Party Cookie, What’s next for Adtech?, DRUM (Jan.

15, 2020), https://perma.cc/TN67-T9FD; Maciej Zawadziński & Michael Sweeney,

Google Chrome to Kill Off Third-Party Cookies: What It Means for AdTech,

CLEARCODE, https://perma.cc/FYA8-DA5Q (last visited Sept. 28, 2020); Ari

Paparo, Google, You Finally Really Did It, ADEXCHANGER (Jan. 14, 2020),

https://perma.cc/EZ7R-SCU9; Weissbrot, supra note 108.

113

. See Seb Joseph, Winners, Losers and Fallout from Google’s Plan to Drop

Cookies, DIGIDAY (Jan. 16, 2020), https://perma.cc/67WG-6RZV (“Unsurprisingly,

Google stands to profit the most from the death of the third-party cookie. In the

absence of third-party cookies’ use with Chrome, the alternative for advertisers

is to use Google’s first-party data within its own tools.”). “For example, exchange

bidding could pass a hashed ID of the Google cookie to its partners and similarly

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

106

For instance, Google Chrome started to track consumers’ web activity di-

rectly at the browser level, obviating Google’s need to rely on cookies at

all for identity information.

114

More recently, Google’s exchange has relied on California privacy

legislation, the California Consumer Privacy Act (“CCPA”), to starve ri-

val exchanges of liquidity and exclude non-Google buying tools from

auctions altogether. If a California consumer visits a website, and selects

the “Do Not Sell My Information” link as mandated by CCPA, Google’s

ad server does not permit non-Google exchanges and buying tools to bid

for publishers’ ad space at all.

115

Google’s approach contrasts with that of

make the same available on a server-to-server basis for various pixel calls. To the

extent that Google is the only platform that makes this available, it would have

the added benefit of effectively forcing everyone (except Amazon) into exchange

bidding if they want user data, and thus into Google’s single auction.” Berry,

supra note 107; McCarthy & Blustein, supra note 112 (“Google is building a moat.

It doesn’t need third-party cookies to track people. It has code live on virtually

every single website and app.”); Weissbrot, supra note 108 (joint statement issued

by industry trade groups the ANA and the 4A’s stating “Google’s decision to

block third-party cookies … would threaten to substantially disrupt much of the

infrastructure of today's internet without providing any viable alternative, and it

may choke off the economic oxygen from advertising that startups and emerging

companies need to survive.”).

114

. Specifically, Google currently always and automatically logs users of

Chrome into the browser itself when users log into a Google service like Gmail

or YouTube. If a user then attempts to log out of Chrome, Google automatically

also logs them out of the Google service, thereby prohibiting users from using a

Google service while not simultaneously being logged into the Chrome browser.

On top of this, Google steers users towards syncing their Chrome user data with

other Google user data and granting Google permission to use this data for the

purpose of selling Google advertising, including “to personalize Search, ads, and

other Google services.” Screenshot of Google Chrome Terms (2020) (on file with

author); see also Matthew Green, Why I’m Done with Chrome, CRYPTOGRAPHY

ENG’G (Sept. 23, 2018), https://perma.cc/ES7H-24DR (discussing how Chrome

steers users to give Google permission over their browsing history data and how

this is deceptive and a “dark pattern”).

115

. Google Ad Manager Help, Integration with IAB CCPA Framework Tech-

nical Specifications, GOOGLE SUPPORT, https://perma.cc/Q9WU-M3AQ (last visited

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

107

the industry trade group—the Internet Advertising Bureau (IAB)—and

rival exchanges.

116

Other exchanges instead send the same bid requests

to all bidders, signaling that a user associated with ad space has exercised

an opt-out under CCPA.

The approach taken by the IAB and rival exchanges mirrors the

broader competition approach that lawmakers take in the equities mar-

ket. Stock exchanges must provide all traders with access to relevant trad-

ing information. Then, firms must comply with industry-specific privacy

rules to safeguard consumer data transmitted during trading.

5. Google-Owned Intermediaries Have a Speed Advantage on

Google’s Exchange

When financial markets migrated to electronic trading, non-discrim-

inatory access to exchanges—specifically, access to data and trading

speed—became critical to competition. The financial traders who had

their computers located physically closer to exchange servers noticed that

they could access relevant trading information first and place orders

faster than the traders whose computers were located further away.

This

led to a speed “arms race.”

117

In the early 2000s, some traders moved their

computers closer to the new computerized exchanges in New York City’s

Financial District to increase trading speeds, before colocation in the

same room with exchanges became the trend.

118

Despite the decentralized

Sept. 28, 2020) (“When restricted data processing is enabled, Google does not call

out to third parties via RTB and the signal would not be propagated further to

other vendors”).

116

. IAB PRIVACY, IAB CCPA COMPLIANCE FRAMEWORK FOR PUBLISHERS &

TECHNOLOGY COMPANIES: VERSION 1.0 (2019); Bid Request to Bidders, supra note

106.

117

. Aaron Lucchetti, Firms Seek Edge Through Speed as Computer Trading

Expands, WALL ST. J. (Dec. 15, 2006), https://perma.cc/8EVV-8GWS (quote of

Tradebot Systems principle David Cummings).

118

. Id. (sharing that trading firm Tradebot Systems Inc. moved their com-

puters from Kansas City to New Jersey and New York to colocate with exchanges

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

108

nature of the Internet, having servers in closer proximity with each other

yields faster response times.

Though trading speed emerged as critical to competition in the early

2000s, it was not until Michael Lewis released a book called FLASH BOYS

in 2014 that a wider audience began to appreciate the implications of

speed and latency with electronic trading.

119

Lewis, a renowned chroni-

cler of Wall Street culture, broke down how speed shaped competition

on stock exchanges and how traders were going to greater and greater

lengths to acquire incremental speed advantages. In an illustration of

how ridiculous the race for speed had become, Lewis recounted how one

enterprising group spent hundreds of millions of dollars drilling through

the Allegheny Mountains of Pennsylvania to lay straighter wires between

the Chicago and New York exchanges to shave off three milliseconds

when trading between the two. Being able to trade a few milliseconds

faster gives traders an edge.

There are speed parallels in advertising markets. Speed shapes com-

petition between buying tools on ad exchanges.

120

Ads trade in millisec-

onds and exchanges limit the amount of time that those intermediaries

have to return bids. If a trading desk, DSP, or other buying tool is too

slow, an exchange will exclude it from the instant auction.

Similarly, if a

buying tool is slow too often, an exchange might “throttle” it.

121

Meaning,

in 2002 and that from 2002 to 2006, about 40 small trading firms and large bro-

kerages moved their computers closer to exchanges); Michael J. McGowan, The

Rise of Computerized High Frequency Trading: Use and Controversy, DUKE L. & TECH.

REV. 24 (2010).

119

. MICHAEL LEWIS, FLASH BOYS: A WALL STREET REVOLT (W. W. Norton

& Company 2014).

120

. See generally supra notes 44-45.

121

. Specifically, the exchange will send the buying tool fewer bid re-

quests, allowing it to participate in only some ad auctions until it improves trad-

ing speeds. See e.g., Latency Restrictions and Peering, GOOGLE DEVELOPERS,

https://perma.cc/C62T-UKAP (last visited Sept. 28, 2020) (“We require that 85

percent of responses be received within the deadline from the perspective of the

trading location and will throttle bidders that cannot consistently achieve this.”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

109

the exchange will exclude it from a percentage of future auctions. Because

the buying tools encounter varying levels of traffic and delay when con-

veying their bids back to exchanges (network latency), they need to leave

enough time to account for such traffic.

122

Thus, as a general matter, bid-

ders that receive bid requests and respond to them fastest have ad-

vantages over those that are slower.

Figure 4: Time Restrictions in Bid Requests

Like the trading firms on Wall Street that benefit from speed ad-

vantages, the Google-owned intermediaries DV360 and Google Ads also

have speed advantages. Google explains to current and prospective cus-

tomers that when DV360 and Google Ads compete against the non-

Google buying tools for the ad space trading in Google’s exchange, they

have a speed advantage from colocation.

123

By colocating DV360 and

Google Ads with Google’s exchange,

the bids submitted by DV360 and

Google Ads can be excluded less often than the bids submitted from bid-

ders that are not colocated.

124

122

. Hasham, supra note 44 (discussing network congestion, Google rec-

ommended buffers, and recommended buffers of 20 milliseconds).

123

. Google Ad Manager Help, supra note 46.

124

. Id.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

110

In fact, colocation may just reduce the frequency of exclusion for the

Google-owned intermediaries from 1-in-4 to zero. According to a post by

Google, DV360 and Google Ads have “no network latency or timeout is-

sues,” meaning that Google’s exchange “always receive bids” from ad-

vertisers that use them.

125

Conversely, in Google’s experience, “latency

issues can prevent advertisers from successfully submitting a bid on up

to 25% of bid requests.”

126

That is, one out of every four bids submitted

to an exchange by a non-colocated buying tool may be disregarded

simply because of speed.

Colocation also provides intermediaries like DV360 and Google Ads

with another distinct benefit. Because they must allocate less time to net-

work travel—the time it takes for the bid request to travel from Google’s

exchange to a non-colocated bidder and back—DV360 and Google Ads

have more time to query additional data about the user to better deter-

mine the value of ad space for sale.

127

That is useful because online ad-

vertisers aim to target particular consumers, drive web traffic, or sell

more widgets, all of which require time to query data, crunch data, and

pick an advertisement that has the highest likelihood of persuading a

user to take a particular action. Time permits DV360 and Google Ads to

bid smarter.

125

. Id.

126

. Id.

127

. See generally Vasavi Pandey, 5 Questions to Ask When Considering Server

Side Header Bidding, INDEX EXCH. (Mar. 1, 2017), https://perma.cc/LP4E-SWMK.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

111

Figure 5: Google Colocation & Unequal Access to Speed Excludes

Bids from Slower Traders

Today, the Google-owned intermediaries are the number-one buying

tools in the market for both small and large advertisers and it is unclear

to what extent this is due to speed advantages on Google’s number-one

exchange. In financial markets, the traders that send and receive signals

to and from financial exchanges fastest and have systemic advantages

over others trading on the same venue.

128

Systemic speed advantages also

128

. See generally AQUILINA ET AL., supra note 24 (providing an overview of

the role of speed and data in the securities market); McGowan, supra note 118

(discussing the shift to computerized trading and the role that superior speed

and data plays with high-frequency trading); Liz Moyer & Emily Lambert, The

New Masters of Wall Street, FORBES (Sept. 2, 2009), https://perma.cc/HVV5-QZBW

(discussing generally how slower traders were excluded from competing against

faster ones); Gregory Meyer et al., How High-Frequency Trading Hit a Speed Bump,

FIN. TIMES (Jan. 1, 2019), https://perma.cc/2BGZ-26XT (discussing ongoing com-

petition around speed and data); Thierry Foucault et al., Equilibrium High Fre-

quency Trading, Int’l Conf. Of the French Fin. Ass’n (Sept. 2011),

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

112

translated to market power. The firms in the early 2000s that first made

speed their strategy—Citadel, Virtu Financial, Getco and Tradebot—

emerged as the eventual market leaders.

129

Even if speed advantages can explain the market power of Google

intermediaries DV360 and Google Ads, Google does offer some form of

colocation to competitors. Since 2013, Google has let competitors colocate

through Google’s proprietary computer infrastructure services division

Google Cloud.

130

In addition to Cloud colocation, Google has partner-

ships with non-Google data centers that guarantee round trip network

https://perma.cc/8ZTN-92DD (studying how differential access to market data

and speed forecloses competition between traders). But see Robert Litzenberger

et al., The Impacts of Automation and High Frequency Trading on Market Quality, 4

ANNUAL REV. FIN. ECON. 58-59 (Oct. 2012) (arguing that discriminatory access to

speed forecloses competition between high frequency traders but not retail in-

vestors, who compete instead on different fundamentals).

129

. Michael Mackenzie, US: High Frequency Trading Dominates the Debate,

FIN. TIMES (Oct. 20, 2009), https://perma.cc/3TUR-9M8E (reflecting that high-fre-

quency traders using speed strategies accounted for approximately 70% of U.S.

equity trading volume in 2009, up from 30% a few years prior); Concept Release

on Equity Market Structure, 75 Fed. Reg. 3594, 3606 (proposed Jan. 21, 2010) (to

be codified at 17 C.F.R. pt. 42) (“Estimates of HFT volume in the equity markets

vary widely, though they typically are 50% of total volume or higher.”); Meyer,

et. al., supra note 128 (providing the same estimate, attributed to the Tabb Group,

and discussing how Virtu Financial leveraged speed early-on to become the larg-

est proprietary trading firm in the U.S.); Jonathan Spicer & Herbert Lash, Who’s

Afraid of High-Frequency Trading, REUTERS (Dec. 2, 2009), https://perma.cc/MRT2-

8A5Y (reporting that trading firms Tradebot and Getco that were amongst the

first to adopt speed as their strategy regularly accounted for 20% of the trading

volume in the U.S. stock market); Kim Janssen, Citadel Trader Testimony Reveals

Little of World That Helped Mint Ken Griffin, CHI. TRIB. (Oct. 28, 2015),

https://perma.cc/W5AR-2734 (discussing how Citadel is a major firm in the high-

frequency trading space).

130

. Open Bidder (Beta), GOOGLE DEVELOPERS (May 19, 2013),

https://perma.cc/DE5G-9ER3; Noam Wolf, Google I/O 2013 - Introducing Open Bid-

der: Leverage Cloud Platform to Reinvent Display Advertising, YOUTUBE (May 19,

2013), https://perma.cc/XB2N-2GTV. Google Compute Engine — Cloud Platform,

GOOGLE (Aug. 5, 2013), https://perma.cc/22VR-QDAY.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

113

travel with Google within 5 milliseconds.

131

It is not clear whether 5 mil-

liseconds of latency make or break a rival’s ability to bid for ad space

trading in Google’s exchange.

At least when it comes to Google Cloud colocation, it appears that

few firms have taken Google up on its offer. One exception is advertising

exchange OpenX, which publicly announced a five-year, $110 million

colocation agreement with Google in January 2019.

132

In discussing the

OpenX deal, Google reiterated the competitive benefits of colocation:

“[colocation] reduced overall latency, especially when sending requests

to Google … It increases the speed that bids come into the OpenX Ex-

change from potential buyers, driving increased revenue potential by al-

lowing publishers to see greater demand for their audiences before an

auction closes.”

133

Despite Google’s arguments for colocating, few other

companies have chosen to follow in OpenX’s footsteps.

It is possible that rivals do not colocate with Google in Cloud because

Google charges high colocation rents or rivals are not comfortable colo-

cating on Google territory. In financial markets, colocation practices are

tightly regulated: procedures and prices have to be approved by the Se-

curities and Exchange Commision (SEC) and the length of cord in colo-

cation facilities has to be exactly the same to ensure non-discriminatory

access. In advertising, the common practice is for ad exchanges and buy-

ing tools to control for some of these neutrality concerns by meeting on

131

. See Choosing Colocation Facility Locations, GOOGLE CLOUD,

https://perma.cc/KPN7-6N38 (last visited Sept. 28, 2020).

132

. George Slefo, Google Cloud, Looking to Invade Amazon’s Turf, Inks Deal

with OpenX, ADAGE (Jan. 24, 2019), https://perma.cc/57ZL-TD7Q (“It's the first

time that Google Cloud has struck a deal with a well-known, or large exchange,

according to both companies.”); Andrew Blustein, OpenX Bets Future Is Serverless

in Move to Google Cloud Platform, DRUM (Jan. 24, 2019), https://perma.cc/TK7Y-

SB3Q (“You have to operate at speed, efficiency, closeness to the publisher and

the demand-side of Google, and if you don't . . . you aren't going to be able to

compete in the adtech space.”).

133

. OpenX: Powering the Future of Advertising with Google Cloud, GOOGLE

CLOUD https://perma.cc/84RY-FPPF (last visited Sept. 28, 2020).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

114

independent ground. For instance, AT&T XNDR hosts its exchange with

intermediaries in independent data centers, including at Equinix—the

same data center that often acts as a neutral arbiter of colocation for stock

exchanges, including the NYSE, Nasdaq, the Chicago Board Options Ex-

change, BATS, and trading firms.

134

Google, unlike Equinix, has conflicts

of interest. Google Cloud’s job is to nullify speed advantages between

competitors, but Google also operates buying tool intermediaries that

benefit from speed advantages over those customers.

When it comes to speed and competition, one problem is that a lack

of transparency makes it difficult for market participants to understand

the true extent that speed favors Google and excludes others. Does colo-

cation really reduce Google Ads’ exclusion from Google’s exchange from

1-in-4 to zero? While Google’s exchange shares centralized auction re-

ports back with publishers, those reports do not include information

about the bids that Google’s exchange excludes due to latency.

135

In other

134

. Datacenters, XANDR CLOUD SERVICES, https://perma.cc/QWT3-7WAK

(last visited Sept. 28, 2020); Matthew Leising & Annie Massa, Inside the Nonde-

script Building Where Trillions Trade Each Day, BLOOMBERG (Apr. 13, 2016),

https://perma.cc/QT9J-F896 (discussing colocation in the securities market).

Other major ad exchanges, including those belonging to Index, OpenX, and Pub-

matic, take the same approach. OpenX Selects Zayo for Data Center Expansion,

ZAYO (Mar. 15, 2016), https://perma.cc/XU54-RVV7; Vasavi Pandey, Tour IX’s

Amsterdam & Frankfurt Data Centers, INDEX EXCH. (July 2, 2018),

https://perma.cc/L7AJ-CVY9; Digital Realty SFR1, INFLECT (2020),

https://perma.cc/J3QV-9AVS (last visited Sept. 28, 2020).

135

. See Bids Data in Ad Manager Data Transfer (Beta), GOOGLE AD MANAGER

HELP (2020), https://perma.cc/R6DN-ZJHV (making no mention of reporting on

bids due to latency); Jason Bigler, Rolling Out First Price Auctions to Google Ad

Manager Partners, GOOGLE https://perma.cc/QL6V-LHXN (last visited Sept. 26,

2020) (stating that new beta market trading files include data from “bids submit-

ted to your auctions”) (emphasis added). The Google Ad Manager reports on

“metrics” and “dimensions” do not include data about bids excluded due to la-

tency or timeouts. See Ad Manager Report Metrics, GOOGLE AD MANAGER HELP,

https://perma.cc/Q7FQ-2FWN (last visited Sept. 26, 2020); Google Ad Manager

Help, Ad Manager Report Dimensions, GOOGLE (2020), https://perma.cc/QZ8Q-

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

115

words, if Google’s exchange hosts a 100 millisecond auction for an ad,

and reports that a Google-owned intermediary won the auction at a price

of a $2 CPM—which stands for cost per one thousand ad spaces (“im-

pressions”)—a publisher would not know if Google’s exchange excluded

a higher $3 CPM bid received just 2 milliseconds late.

In financial markets, specific investigations and enforcement actions

have been critical to discovering speed problems, enforcing fair access

rules, and protecting competition.

For example, in 2012, the NYSE settled

charges levied by the SEC for giving some trading firms access to the

market single digit milliseconds to multiple seconds before others.

136

Fol-

lowing FLASH BOYS’ release in 2014, the FBI, the SEC, and the securities

enforcement division of the Office of the New York Attorney General

(NYAG) opened investigations into whether other exchanges were in vi-

olation of existing fair access rules.

137

The second-largest public exchange,

BATS, soon settled charges brought against them for providing some

traders with superior access.

138

In advertising, a lack of transparency by

V6DT. The following NetworkBackfillBids report states it includes bids catego-

rized as “Other,” but does not mention latency; the prior documentation suggests

that “Other” would exclude bids due to latency. See also Ad Manager Data Transfer

Reports, GOOGLE AD MANAGER HELP, https://perma.cc/8QSV-YBQF (last visited

Sept. 28, 2020) (describing the various market trading reports Google offers). The

NetworkBackfillBids report contains data about Google’s market trading activity

and pulls from winning bids, bids excluded because they were lower than the

auction “Floor,” bids that were “Outbid,” and bids excluded due to “Other,” or

those Google calls “lost for another reason, such as URL filtering.” Google Ad

Manager Help, Bids Data in Ad Manager Data Transfer (Beta), GOOGLE ,

https//:perma.cc/2HN6-AVY4 (last visited Dec. 2, 2020).

136

. N.Y. Stock Exch. LLC, supra note 48.

137

. See generally Michael Lewis, Michael Lewis Reflects on His Book Flash

Boys, a Year After It Shook Wall Street to Its Core, VANITY FAIR (Mar. 12, 2015),

https://perma.cc/U92W-DKLL (summarizing onset of regulatory scrutiny after

the release of FLASH BOYS.

138

. N.Y. Stock Exch. LLC, supra note 48 (two exchanges owned by BATS

Global settle their charges for not adequately disclosing how their order types

worked and for giving information about order types to some traders but not

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

116

trading intermediaries, combined with a lack of regulatory oversight,

may permit speed exclusion problems to go under the radar.

Finally, it is important to note how unequal trading speeds can neg-

atively affect the efficiency of the trading market, in addition to leading

to market concentration. Unequal access to speed can operate to line the

pockets of the middlemen. The following hypothetical provides one ex-

ample of how this could happen. When Google’s exchange excludes rival

intermediaries for being too slow, the Google-owned intermediary

Google Ads would have been able to acquire publishers’ ad space for

lower prices. But it is not clear whether the small mom-and-pop adver-

tisers using Google Ads benefitted from such lower prices. Instead, the

setup could have permitted Google as the intermediary to engage in a

wider arbitrage—keeping a larger difference between the price it sells an

ad to a Google Ads advertiser and the price the ad space actually clears

for on Google’s exchange.

139

Like in finance, the mechanics of highly-so-

phisticated electronic trading markets and their accompanying speed

others); In re Ubs Sec. LLC Respondent., SEC Admin. Release No. 9697, 2015 WL

179551 (Jan. 15, 2015), https://perma.cc/7KXP-EMJ5 (UBS settles its charges for

violating Regulation NMS by allowing some traders to place orders in small in-

crements, which allowed some orders to jump ahead of others).

139

. Though not completely on point, there is one case in which an inter-

mediary in the event ticket market sequenced buyers through two consecutive

marketplaces that it owned and was thereafter charged with violating Section 5

of the FTC Act. LiveNation, the company that owns the Ticketmaster market-

place, also operates a second marketplace called TicketsNow. In the late 2000s,

Ticketmaster automatically re-directed consumers looking to purchase tickets on

Ticketmaster to the TicketsNow marketplace. TicketsNow listed the same tickets

at higher mark-ups, which permitted the parent company to obtain higher com-

missions on these trades. The FTC brought charges against Ticketmaster and

TicketsNow for steering consumers from one marketplace to another, alleging

that this conduct was deceptive and unfair and constituted a violation of Section

5 of the FTC Act. The two companies settled and agreed to pay consumers the

difference between what they paid to purchase the tickets in the TicketsNow

marketplace and what they would have paid in the Ticketmaster marketplace.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

117

races may just wash over the heads of the least sophisticated market par-

ticipants.

B. Discriminatory Routing of Orders and More Speed Races

1. Routing Orders to Google’s Exchange and Owned Properties

While exchanges can distort competition between intermediaries by

sharing data and speed in an asymmetric fashion, the intermediaries can

distort competition by routing buy and sell orders in a discriminatory

manner. In the equities trading market, we temper competition and wel-

fare problems associated with this conduct by identifying and managing

intermediary conflicts of interest. Your broker cannot steer you to pur-

chase stock in which they have an underlying financial interest. Neither

can the broker steer you to purchase shares of stock from the broker’s

exchange if the same shares trade on a rival’s exchange for lower prices.

Google takes advantage of the lack of conflicts of interest oversight in

advertising markets to distort competition in similar ways. For example,

when this author opened a Google Ads account, Google Ads required the

author to start a Google Search campaign in order to complete the ac-

count setup process.

140

In other words, Google Ads does not merely steer

See Ticketmaster and TicketsNow Settle FTC Charges of Deceptive Sales Tactics, Re-

funds for Springsteen Concertgoers Provided; FTC Warns Other Ticket Resellers, FED.

TRADE COMM’N (Feb. 18, 2010), https://perma.cc/LM9R-947W. Separately, in the

foreign exchange market, which is not subject to the rules and regulations that

apply to the equities market, the DOJ and SEC brought civil charges against the

asset manager State Street for defrauding investors by charging hidden mark-

ups and not properly disclosing how its electronic platform for trading U.S.

Treasuries worked. See David Lynch, State Street Agrees $35m Fraud Settlement

with SEC, FIN. TIMES (Sept. 7, 2017), https://perma.cc/Q8QR-4Z5K.

140

. How Google Ads Requires Advertisers to Purchase Search Ads (2020)

(on file with author). Additionally, according to a recent Wall Street Journal in-

vestigation, Google Search also steers users to YouTube videos. Sam Schechner

et al., Searching for Video? Google Pushes YouTube Over Rivals, WALL ST. J. (July 14,

2020), https://perma.cc/P59U-RQCR.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

118

advertisers to Google Search, it sometimes requires them to purchase

Google Search ads before they can bid on display space belonging to pub-

lishers such as The Seattle Times.

Exacerbating steering and tying conduct is the fact that many adver-

tisers must use a Google-owned buy-side intermediary to purchase

Search or YouTube ad space, properties many advertisers cannot forgo.

Since 2015, advertisers that want to purchase YouTube ads must use

DV360 or Google Ads because Google withholds YouTube ad space from

rival buying tools.

141

It is less clear whether and to what extent Google

withholds Search inventory from rivals. In January 2012, the Federal

Trade Commission found that Google restricted third-party tools from

selling Google Search ads and that such restrictions were anticompeti-

tive.

142

But the FTC ultimately declined to file suit, and Google loosened

141

. Jeff Fraser, Google Cuts out Third-Party DSPs from YouTube Inventory,

MARKETING (Aug. 7, 2015), https://perma.cc/N6JD-SPYT (explaining how Google

withdrew YouTube inventory from third-party buying tools).

142

. Staff report from the competition bureau stated the following: “Staff

has investigated whether Google has employed anticompetitive contractual re-

strictions on the automated cross-management of advertising campaigns.

Google’s main rival (Microsoft) has alleged that Google is denying Microsoft crit-

ical scale by employing these restrictions, and thus impairing Microsoft’s ability

to compete effectively in the markets for general search and search advertising.

We conclude that these restrictions should be condemned under Section 2 be-

cause they limit the ability of advertisers to make use of their own data, and as

such, have reduced innovation and increased transactions costs among advertis-

ers and third-party businesses, and also degraded the quality of Google’s rivals

in search and search advertising.” The FTC Report on Google’s Business Practices,

WALL ST. J. (Mar. 24, 2015). David Drummond, The Federal Trade Commission

Closes Its Antitrust Review, GOOGLE (Jan. 3, 2013), https://perma.cc/XUJ5-B9W4

(explaining how Google would start permitting third-party tools that connect to

the AdWords API to “mix and copy campaign ad data,” as advertisers are per-

mitted to do when they use AdWords directly); Greg Sterling, From Praise to Out-

rage: Reactions to Google’s Antitrust Settlement, SEARCH ENGINE LAND (Jan. 4, 2013),

https://perma.cc/SMU8-5K2V (summarizing how Google’s changes helped third-

party tools like Marin Software and Kenshoo better sell Google Search ads). Sep-

arately, for Google’s share of the US search market, see Search Engine Market Share

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

119

its restrictions. The competition agency in the U.K. recently re-investi-

gated the matter.

143

On top of pressuring advertisers to use Google’s buy-side tools,

Google pressures advertisers to use just one tool at a time (i.e., to single

home).

144

Because Google scrambles DoubleClick IDs for everyone but

Google, advertisers that use more than one buying tool at a time risk in-

advertently bidding against themselves in exchanges. Buying Tool 1

might bid on user 12345, while Buying Tool 2 might try to outbid the first

for user 09876, without the two knowing that user 12345 and 09876 are in

fact the same person. By using two tools at once, advertisers can drive

their acquisition prices up. As a result, advertisers have pressure to single

home, and if they want to purchase advertising on YouTube, they should

single home using DV360 or Google Ads.

According to Google, conflicts of interest should lead to efficiencies

rather than distortions in the market: “the combination of Google’s search

business and its vertical ad tech integration should give it incentives to

in 2020, OBERLO (2020), https://perma.cc/27VF-7DXY (stating that Google has an

88+% share of U.S. search queries and a 92+% share of worldwide search queries).

143

. The competition agency in the UK, Competition Markets Authority,

is investigating non-Google buying tools’ access to Search inventory. The CMA

FINAL REPORT stated that: “Transaction costs faced by advertisers in using multi-

ple search engines appear likely to benefit Google and create a barrier to entry

for smaller rivals. Most of the larger advertisers we contacted did not suggest

there was a difficulty in multi-homing across multiple search engines and used

tools such as SA360, or those provided by third parties such as Marin and

Kenshoo, to do so. However, a minority of the larger advertisers who responded

to our questionnaire choose to single-home as Google had a sufficiently large

market share and reach to meet all their needs. Based on evidence from the qual-

itative research set out in Appendix N, many smaller advertisers choose to sin-

gle-home, using the simple interface provided by Google Ads and benefiting

from its broad reach.”CMA FINAL REPORT, supra note 7, at 227.

144

. For a discussion of the risks of bidding against oneself when targeting

individual users in advertising auctions, see Jaysen Gillespie, [RESEARCH] Why

You Should Only Use One Retargeter, CRITEO (July 1, 2019), https://perma.cc/FWK6-

PSZC; Brian O’Kelley, DSP User: In What Situations Am I Bidding Against Myself?,

QUORA, https://perma.cc/2CRT-4J7F.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

120

balance the interests of all ecosystem participants . . . .”

145

However, the

data tell a different story.

146

Google’s public filings generally show how

the split in the allocation of advertising revenues between Google and

non-Google properties has widened almost every year since 2004.

147

2007, the share going to Google properties increased to 64%, in 2008 to

68%, eventually to 71% (2011), then 75% (2014), 77% (2015), 80% (2016),

81% (2017), and 82% (2018). It increased again in 2019 with just 16% of

the $134 billion dollars that advertisers spent through Google going to

the 2+ million non-Google properties, like The Post, The Register, Le Monde,

or The Times of India, that also sell their ad space through Google’s ex-

change and buying tools.

148

Early 2020 data suggest this squeeze will con-

tinue. In the midst of a global pandemic, Q1 2020 data reveals the share

going to news publishers has further declined to 15%.

149

Google Search in particular accounts for 73% of Google advertising

revenues, though it is not clear why.

150

One cannot definitively attribute

revenue growth to usage growth because Google stopped reporting

Search usage and query growth.

151

Some external estimates even suggest

145

. Bitton & Lewis, supra note 15, at 12 ( “Google’s presence across the

buy and sell-side of ad tech also gives it incentives to promote efficiency tradi-

tionally associated with vertical integration, including the reduction of double

marginalization . …”).

146

. The fact that these types of competition problems arise again-and-

again in electronic trading markets also pushes back on that vertical integration

hypothesis. A similar issue came up in the market for electronically traded air-

fares. When the airline-owned electronic marketplaces for airfares engaged in a

similar pattern of withholding then steering behavior in the 1970s, lawmakers

intervened to put a stop to it. See infra note 209.

147

. Google Inc. (2005-2016) and Alphabet Inc. (2017-2020), supra note 10.

148

. See Alphabet Inc. (Feb. 4, 2019), supra note 10; About Targeting for Dis-

play Network Campaigns, GOOGLE ADS HELP, https://perma.cc/U24D-PRRW.

149

. Alphabet Inc. (Feb. 3, 2020), supra note 10.

150

. Bitton & Lewis, supra note 15, at 10; Alphabet Inc. (Feb. 4, 2019), supra

note 10, at 29.

151

. See Danny Sullivan, Google Now Handles at Least 2 Trillion Searches per

Year, SEARCH ENGINE LAND (May 24, 2016), https://perma.cc/LT4T-HCTQ.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

121

growth post-2012 was stagnant or slow.

152

To shareholders, Google has

been generally attributing revenue increases to “interrelated factors,” in-

cluding “growth in advertiser activity.”

153

But as this author’s experience

with Google Ads shows, Google has the ability to generate advertiser ac-

tivity through steering and tying. It is impossible to knowwhether

Google also quietly steers demand by the way that it allocates the use of

data when making ad-matching decisions.

154

152

. See Danny Sullivan, Google Still Doing at Least 1 Trillion Searches per

Year, SEARCH ENGINE LAND (Jan. 16, 2015), https://perma.cc/9URP-VU7B (sum-

marizing that Google was issuing the same daily search number of 3 billion in

2015 that it disclosed in 2012, that comScore numbers showed a 5% gain from

12/2012 to 12/2014, and that the analysis by journalist Charles Arthur suggests

low search usage from mobile devices).

153

. Alphabet Inc. (2019), supra note 10, at 30 (attributing revenue in-

creases to “interrelated factors including increases in search queries resulting

from ongoing growth in user adoption and usage, primarily on mobile devices,

continued growth in advertiser activity, and improvements we have made in ad

formats and delivery”).

154

. Google charges advertisers by user conversion (i.e., when a user takes

a specific action). Thus, when Google knows that a user is a conversion target, it

can distort competition between sellers by displaying an ad to that user when

they are on a Google property, as opposed to a third-party property. In financial

markets, trading intermediaries owe fiduciary or best interest duties to custom-

ers, which requires them to allocate trading decisions in a fair and equitable man-

ner. For example, intermediaries cannot engage in “trade allocation,” the alloca-

tion of favorable trades to certain accounts and less favorable trades to other

accounts.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

122

Figure 6: Share of Google Ad Revenues Going to Google vs. Non-

Google Properties 2004-2019

155

In addition to steering advertisers to Google properties, Google’s

trading intermediaries steer buyers and sellers to execute orders on

Google’s exchange. Until 2016, on the buy-side, Google Ads only routed

advertisers’ bids to Google’s exchange.

156

Meaning, if the local dry cleaner

used Google Ads to bid on ad space belonging to The Register, Google

Ads did not route the dry cleaner’s bids to competing exchanges belong-

ing to AppNexus, Index, and Pubmatic, where The Register inventory

may have been trading for lower prices. According to a claim by the CEO

of rival exchange AppNexus, Google’s enterprise buying tool DV360 en-

gaged in similar conduct.

157

Today, though both DV360 and Google Ads

155

. See Google Inc. (2005-2016) and Alphabet Inc. (2017-2020), supra note

10. Note: Google acquired YouTube during Q4 of 2006. “Google Properties” also

includes properties such as Gmail. See supra notes 10-12.

156

. DoubleClick Ad Exchange, supra note 34.

157

. Sarah Sluis, AppNexus CEO Brian O'Kelley on Waging a Price War,

ADEXCHANGER (Nov. 9, 2017), https://perma.cc/8KHY-MGX6 (then-CEO of

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

123

technically route orders to non-Google exchanges, they would still dis-

proportionately route orders to Google’s because of the speed and infor-

mation advantages discussed in Part III.A.

158

Google’s ad server, the dominant intermediary on the sell-side, also

steers orders to Google’s exchange. From 2010 through 2018, the ad

server restricted websites from transacting with advertisers through non-

Google trading venues in two ways.

159

First, Google prohibited websites

from routing their ad space to multiple venues at the same time,

160

the

Google rival exchange AppNexus stating that “Google is quietly allocating de-

mand toward AdX, but without making any rationale to the buyer as to why. It’s

bad for the buyer. It’s bad for the publisher”).

158

. GOOGLE, ONLINE PLATFORMS AND DIGITAL ADVERTISING COMMENTS ON

THE MARKET STUDY INTERIM REPORT 11 (2020), https://perma.cc/CL3R-ZL4M

(Google claiming that it is “wrong” to think that Google Ads and DV360 prefer

Google’s exchange). But see Google Ad Manager Help, supra note 46 (explaining

that publishers benefit from greater “auction pressure” and higher demand for

their inventory in Google’s exchange when advertisers use DV360 and Google

Ads because of user ID and speed reasons).

159

. For a thorough explanation on how Google’s ad server restricted com-

petition in the exchange market, see Damien Geradin & Dimitrios Katsifis, An

EU Competition Law Analysis of Online Display Advertising in the Programmatic Age,

15 EUR. COMPETITION J. 55 (2019).

160

. The process of routing advertising impressions to exchanges in se-

quence was also called “waterfalling” or “daisy-chaining.” In the online adver-

tising market’s early days, before the invention of real-time impression-by-im-

pression bidding through exchanges, it was rational for the ad server to provide

publishers with sequential routing because publishers sold their ad space to ad-

vertisers and through ad networks for fixed CPM prices. However, once the mar-

ket structure migrated to real-time impression-by-impression bidding in the late

2000s, sequential routing stopped being rational and started to operate as a re-

striction on exchange competition. A rough parallel today might be if a financial

intermediary prohibited financial exchange competition based on a legacy prac-

tice designed in the days of floor-based trading. See Gavin Dunway, What Is the

Waterfall?, ADMONSTERS (Feb. 23, 2017), https://perma.cc/Q8BR-2M9V (further

discussing history of waterfalling); Bitton & Lewis, supra note 15, at 13 (further

explaining how waterfalling has its origins in the early 2000s and was used by

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

124

way that shares of stock, airline tickets, and sports tickets trade in multi-

ple marketplaces at once.

161

Pitting ad exchanges against each other

would let the demand in each exchange naturally drive prices up. In ad-

vertising, however, websites had to go through their ad server, which

was most commonly Google’s (90+% market share).

162

Google’s ad server

let publishers route their inventory to one exchange at a time, in sequen-

tial order—requiring a site like The Register to route ad space first into

Exchange A, then to Exchange B, only if Exchange A passed on the op-

portunity.

A second way that Google’s ad server restricted competition between

exchanges was by incentivizing websites to route their ad space to

Google’s marketplace first: if a site did, the server permitted a limited

publishers selling inventory for fixed prices); George Slefo, Google’s Exchange Bid-

ding Makes Its Debut: Here’s What You Need to Know, ADAGE (Apr. 4, 2018),

https://perma.cc/E5HH-N6NZ (reporting on Google’s official launch of “Ex-

change Bidding,” which permitted publishers to route their ads from their

Google ad server to multiple exchanges at the same time).

161

. Stocks trade on public exchanges, such as the NYSE and Nasdaq, and

on 30+ Alternative Trading Systems (ATSs).

162

. Complaint at 42, Devaney v. Google, No. 5:20-cv-04130 (N.D. Cal.

June 22, 2020); Keach Hagey & Vivien Ngo, How Google Edged Out Rivals and Built

the World’s Dominant Ad Machine: A Visual Guide, WALL ST. J. (Nov. 7, 2019).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

125

form of exchange competition (that highly favored Google) in return.

163

Most publishers did exactly that to maximize yield on their inventory.

164

163

. Specifically, if publishers routed their ad space to Google’s exchange

through a feature Google called “Dynamic Allocation,” Google’s ad server per-

mitted Google’s exchange–and only Google’s exchange–to compete against the

“floor prices” that publishers statically set for rival exchanges. To be clear, those

floor prices were required by the ad server to be manually set by publishers. Pub-

lishers commonly set those values to equal what an exchange historically cleared

ad space for on average. Static floor prices post-2010 made little sense, since ex-

changes could return a live informed bid for particular impressions if Google’s

ad server permitted them to do so. Such was Google’s ad server’s official setup

from 2010 through 2018, though the market migrated to real-time impression-by-

impression bidding in the late 2000s. The subject of Google’s ad server’s contin-

uance of this Dynamic Allocation conduct post-2010 is a point of particular con-

tention. In a recent filing to the Australian competition regulator, Google de-

fended its conduct on the basis that it was DoubleClick, not Google, that

introduced Dynamic Allocation before Google’s acquisition of the company in

2008. It pointed to the fact that a page on DoubleClick’s website from 2007 dis-

cusses Dynamic Allocation in conjunction with a DoubleClick exchange. How-

ever, the public record shows that DoubleClick at the time had neither officially

launched an exchange nor put to practice this Dynamic Allocation conduct. The

competitive effects of any dynamic allocation conduct in 2007 would also have

been substantively different, since plans for a DoubleClick exchange did not in-

clude real-time impression-by-impression bidding. Additionally, a former Dou-

bleClick employee shared with me (uncorroborated) that DoubleClick had de-

veloped plans for an exchange to spur acquisition interests and that such

interests did follow The New York Times’ coverage of DoubleClick’s exchange

plans. Telephone Interview with former DoubleClick employee (Feb. 14, 2020).

Based on this person’s account, DoubleClick specifically pitched the sale of the

DoubleClick ad server as something that a firm in the intermediary trading mar-

ket could leverage and claimed that Google executive Neal Mohan had a spread-

sheet that specifically modeled rent extractions through such leveraging during

merger negotiations. Note: that narrative may counter the FTC’s assessment of

the merger and its conclusion that Google would not be able to leverage the Dou-

bleClick ad server to preference itself in the ad trading intermediation market.

See Maximizing Advertising Revenues for Online Publishers, GOOGLE (2010),

https://perma.cc/3VE9-YWPF (original announcement describing how Google’s

ad server let Google’s exchange compete in real-time against other exchanges’

static prices); Google Ad Manager Help, Dynamic Allocation, GOOGLE (2020),

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

126

Discriminatory routing of order flow from the sell-side and buy-side

to Google’s exchange jumpstarted Google’s entry into the exchange mar-

ket. Google entered a crowded market in 2009, and by 2013, Google’s ex-

change was trading the most volume. Rival exchanges, starved of liquid-

ity, fell quietly in parallel.

But this practice and resulting competition problem is neither new

nor novel. Operating parallel to Google’s timeframe, in the equities mar-

ket, the broker dealer Barclays started preferentially routing its clients’

stock orders into Barclays’ recently launched “dark pool”—a trading

venue where parties can trade with each other anonymously.

165

By 2013,

https://perma.cc/929B-GDRA (further explaining how this tactic worked and

continues to work in limited form); Michalis Pachilakis et al., No More Chasing

Waterfalls: A Measurement Study of the Header Bidding Ad-Ecosystem, ASS’N

COMPUTING MACHINERY (2017), https://perma.cc/RSZ4-M4BC (summarizing that

Google’s exchange historically competed against other exchanges’ “average price

of [] past purchases”); Sarah Sluis, The Rise of “Header Bidding” and the End of the

Publisher Waterfall, ADEXCHANGER (June 18, 2015), https://perma.cc/MPY6-6X9F

(further explaining how Google’s ad server preferentially routed advertising in-

ventory to Google’s exchange); Bitton & Lewis, supra note 15, at 15-16, 25 (de-

fending Dynamic Allocation preferencing conduct as something invented by

DoubleClick and not Google); Louise Story, DoubleClick to Set Up an Exchange for

Buying and Selling Digital Ads, N.Y. TIMES (Apr. 4, 2007), https://perma.cc/2LBZ-

5PQY (reporting on DoubleClick’s plans to launch an exchange); FTC Closes In-

vestigation, supra note 80; supra Part III.A.1. Additionally, in 2015, Google’s ad

server let Google’s exchange access and sell additional high-value website adver-

tising inventory and did not grant non-Google exchanges access to the same.

John McDermott, Google Sweetens Deal for Publishers with Dynamic Price Floors,

DIGIDAY (Mar. 5, 2015), https://perma.cc/8PKK-CE7L (reporting that, according

to a Google customer support page, Google’s ad server routes publishers’ direct-

sold impressions only to Google’s ad exchange).

164

. Barry Levine, MarTech Landscape: What Is Header Bidding — and Why

Should Publishers Care?, MARTECH TODAY (Dec. 21, 2015), https://perma.cc/NQC3-

NAFF ( “Many of the site publishers utilizing Google’s DFP ad server employ a

setting that allows its Ad Exchange (AdX) to outbid any of the winning waterfall

bidders by even a penny per CPM, because AdX gets the last bid. This is sup-

posed to maximize yield, but it also puts AdX in a privileged position.”).

165

. Amended Complaint, New York v. Barclay Inc., No. 451391/2014

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

127

Barclays’ specialized trading venue became one of the top two in the U.S.

In the equities market, however, we protect the market from these types

of competition problems by prohibiting this type of steering conduct. The

investment bank eventually paid $70 million dollars in fines to settle

charges brought by the NYAG and the SEC.

166

In addition to distorting competition in the exchange market, it is im-

portant to note how routing order flow to Google’s exchange may have

also resulted in welfare problems. For example, by blocking a publisher

from routing its ad space to exchanges simultaneously, Google’s ad

server could sell the publisher’s space to, for example, the Google-owned

intermediary Google Ads for a high bid of a $10 CPM, even though de-

mand from advertisers in rival exchanges might have otherwise pushed

this clearing price up.

167

In other words, by excluding competition, a pub-

lisher like The Post might have made less advertising revenue than it oth-

erwise would have. Did these lower clearing prices benefit the advertis-

ers buying through Google Ads? Again, the answer to this question

remains unclear. Because Google arbitrages between two Google-con-

trolled marketplaces, lower prices may have instead permitted Google as

the intermediary to retain a wider spread.

168

2. Market Creates Invention to Circumvent Routing Restrictions

(N.Y.S. 2d 2015), [hereinafter Barclay Inc.] (discussing Barclays’ preferential or-

der routing practices, market share, and alleging Barclay made material misrep-

resentations to customers about how it operated its dark pool).

166

. Barclays, Credit Suisse Charged with Dark Pool Violations, U.S. SECURITIES

AND EXCHANGE COMMISSION (Jan. 31, 2016), https://perma.cc/54KJ-LWWD (an-

nouncing settlement of NYAG and SEC claims against Barclays).

167

. Note that bids from non-Google intermediaries could be excluded

from participating in Google’s exchange due to latency reasons discussed in Part

III.A.

168

. See supra note 15.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

128

and Set Own Speeds

Of all Google’s steering and discriminatory routing conduct, the one

that drew the most ire in the industry was Google’s ad server not letting

publishers simultaneously route their ad space to more than one ex-

change at once. When exchanges like the NYSE reflected a huddle of trad-

ers on a street “curb” or trading floor, it was a physical feat to trade on

more than one exchange at a time. That was also why the NYSE was a

natural monopoly for over a century. But the advent of electronic com-

munications broke down these walls and ushered in a watershed of elec-

tronic exchange competition. Shares of Microsoft stock now trade on doz-

ens of electronic trading venues at the same time, pitting exchanges

against each other to fight for market share and drive down trading costs.

Why couldn’t ads?

At the tail end of 2014, publishers started to use a technological hack

to get around Google routing restrictions and push the advertising mar-

ket to function more like other sophisticated electronic trading mar-

kets.

169

The invention, commonly called “header bidding,” required web-

sites to insert a piece of JavaScript code into the header section of their

webpages. As soon as a user visited a page, the JavaScript would route

the website’s ad space to multiple exchanges simultaneously, before

169

. The industry has copiously covered header bidding and how this

workaround was designed to circumvent Google restrictions specifically. See

Header Bidding: Holistic Ad Serving Is Here, AD OPS INSIDER (June 1, 2015),

https://perma.cc/WR2F-WHCN; Sluis, supra note 163; Ross Benes, ‘An Ad Tech

Urban Legend’: An Oral History of How Header Bidding Became Digital Advertising’s

Hottest Buzzword, DIGIDAY (June 16, 2017),https://perma.cc/97F8-F3ZJ/; Michal

Wlosik, Client-Side vs. Server-Side Header Bidding: Pros and Cons, CLEARCODE (Apr.

17, 2019), https://perma.cc/6N8Q-NJJZ (stating that header bidding “originally

emerged as a novel solution to remedy the inefficiency, and Google’s preference,

of its own ad exchange (AdX), during the waterfall ad-buying process . . . .”); Sa-

rah Sluis, The Year Header Bidding Went Mainstream, ADEXCHANGER (Dec. 27,

2016), https://perma.cc/E6AW-QNGZ/ (“After all, the priority that Google gave

its Ad Exchange service kicked off the whole header bidding trend.”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

129

Google’s ad server had time to kick in and prevent them from doing so.

170

The invention permitted a site like The Post to start routing its inventory

to a dozen exchanges at once, instead of routing them to Google’s ex-

change first, something The Post now does.

171

The original version of header bidding (called “client-side”) also pro-

vided buyers and sellers with advantages when it came to speed. Specif-

ically, it let publishers set the time that exchanges have to respond with

their winning bids, which publishers started lengthening to account for

more bids from slower traders.

172

Setting speeds several seconds long

would cause users to load pages without ads or cause users to leave the

page while impatiently waiting for pages to load. But giving exchanges

incrementally more time—a fraction of a second that users might not no-

tice—excluded fewer bidders from auctions and resulted in more and

higher bids returned.

173

Typically, publishers set these exchange timeouts to 500-1000 milli-

seconds (one-half to one second), but the default for the leading header

170

. See Vidakovic, supra note 83; Pachilakis et al., supra note 163.

171

. # ADS.TXT FILE FOR WASHINGTONPOST.COM, https://perma.cc/46UY-

23SL (last visited Nov. 17, 2020) (an initiative from the Interactive Advertising

Bureau Technology Laboratory listing the companies authorized to sell The

Washington Post inventory).

172

. With client-side header bidding, websites use a “wrapper” to set a

“universal timeout,” which is the time limit within which all exchanges compet-

ing for ads must submit a bid. See, e.g., Basic Prebid.js Example, PREBID (2020),

https://perma.cc/59DN-RMSR.

173

. See, e.g., Lynne Johnson, Publishers Should Experiment with Timeouts to

Bring in Higher Bids, ADMONSTERS (Sept. 26, 2019), https://perma.cc/M4YP-FLDR;

Index Exchange Introduces Adaptive Timeout, Incorporating Machine Learning into

Header Bidding, GLOBENEWSWIRE (May 20, 2019), https://perma.cc/2HB3-NLX4

(explaining the correlation between auction speed and revenue: “By adaptively

modifying the timeout, the feature ensures the maximum number of bids make

it to the publisher, thereby increasing revenue, while improving user experience

in each unique condition . . . .”); How to Optimize Header Bidding Setup, PREBID,

https://perma.cc/NG7G-HWKC (“Working with 10 bids (orange) makes incre-

mentally more money as the ad server waits longer. But the 5 bids revenue plat-

eaued.”).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

130

bidding vendor PreBid is set to 3000 milliseconds (three seconds).

174

One

report based on data from 300+ publishers recommended publishers set

their timeouts to at least 1157 milliseconds, finding publishers routinely

exclude higher bids from slightly slower traders when they set stricter

timeouts.

175

In a roundabout way, this setup empowered sellers of ad

space to neutralize any speed advantage that Google Ads and DV360 had

in Google’s exchange due to colocation.

The client-side header bidding invention also increased trading

transparency, permitting market participants to better understand how

speed and competition go hand-in-hand. Publishers could now monitor

how many milliseconds it took exchanges to return bids.

176

If a publisher

set its timeouts at 800 milliseconds, and saw that this timeout excludes

higher bids from exchanges that arrived at 820 milliseconds, the pub-

lisher might re-set its timeouts to 820 milliseconds to increase competi-

tion for its ad space and drive prices up.

174

. Publisher API Reference, PREBID (2020), https://perma.cc/P23L-2824.

Though 3,000 milliseconds (three seconds) was a long time for users, a study of

35,000 websites showed that this amount of time in fact captured 90% of bids

returned on average. Note: an observed overall timeout is different than the ex-

change timeout set by the publisher. Pachilakis et al., supra note 163 (showing

that 600 milliseconds was an observed median timeout and 3,000 milliseconds

was a common timeout that captured 90% of bid responses). Ultimately, publish-

ers tested different millisecond settings, weighing their interests in readership

and revenue to find an optimal balance for their own business. See generally John-

son, supra note 173; Prebid.js FAQ, PREBID (2020), https://perma.cc/9AYE-BR5E

(summarizing that publishers should adjust speeds to fill inventory “at the high-

est CPMs without adding too much latency in the process”).

175

. Johnson, supra note 173; MonetizeMore Publisher Metrics August 2019,

MONETIZEMORE (2019), https://perma.cc/V49A-2FAA.

176

. Vidakovic, supra note 83 (showing a screenshot of this transparency);

see also Appendix B.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

131

Google initially argued that header bidding disadvantaged websites

because it was too slow.

177

But the advantages that the invention pro-

vided—transparency, control over latency restrictions, and the ability to

route inventory to multiple exchanges simultaneously—led publishers to

rapidly adopt it. In 2015, within months of its debut, header bidding was

already the hot topic at industry conferences.

178

By 2016, approximately

70% of major U.S. publishers adopted it.

179

By early 2018, 80% of news

publishers were on board.

180

The effects of unleashed competition reverberated quickly through

the industry. More competition from exchanges and slower bidders

177

. Levine, supra note 164 (discussing an AppNexus executive’s rebuttal

that header bidding decreased latency since 78% of header bidding transactions

concluded within 200 milliseconds, whereas only 12% of sequential auctions

did).

178

. Id. (“Next to ad blocking, header bidding is one of the hottest topics

among digital advertisers and website publishers.”). This market development

was widely viewed in the industry as presenting a viable mechanism to challeng-

ing Google’s hold in the exchange and buying tool markets, as well as Google’s

ability to possibly extract wide intermediary spreads. See generally Ciaran

O’Kane, Why 2016 Was the Beginning of the End of the Current Ad Tech Cycle,

EXCHANGEWIRE (Dec. 23, 2016), https://perma.cc/GPX7-EM6G

(“Header bidding

was the ultimate assault on the ad server. It caught Google completely off guard.

Its money pit – dynamic allocation – was being challenged through its own

tech.”); Sluis, supra note 169 (where an industry executive notes how fast Google

took note of and responded to header bidding).

179

. See generally Sluis, supra note 169; Wlosik, supra note 169 (discussing

why header bidding became the new industry standard); Header Bidding Gains

Momentum, Drives Up Publisher Ad Revenue, BUS. INSIDER (May 3, 2016),

https://perma.cc/YFF2-KBVJ (executive with AppNexus sharing that nearly 70%

of top publishers in the U.S. adopted header bidding); A. Guttmann, Share of Pub-

lishers Doing Header Bidding in the United States as of November 2017, STATISTA (Oct.

1, 2017), https://perma.cc/J5HY-8QDL (showing 71.6% of publishers in the study

in the U.S. adopted header bidding); U.S. Websites that Use Header Bidding, by Cat-

egory, Sep 2017-May 2018, EMARKETER (2017), https://perma.cc/KF2V-ZH8H

(showing that as of May 2018, 81% of news publishers in the US were using

header bidding).

180

. U.S. Websites that Use Header Bidding, supra note 179.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

132

caused publishers’ ad revenue to soar overnight, often by 40-70%.

181

prices on one entertainment and news media publisher increased by

50%.

182

Another publisher reported a daily revenue increase of 40%

within the first day of routing to multiple exchanges and an increase of

100% in the longer-term.

183

Other sites echoed revenue increases of over

100%.

184

What this all meant was that by suppressing competition from

non-Google exchanges, Google had cost publishers real revenue, some-

times significant sums. This was no small development for publishers

large and small reeling from the demise of print advertising and migrat-

ing to sophisticated electronic trading.

3. Google AMP Speed Protocol Restricts Trading Through Non-

Google Venues

With stock trading, speed races have shifted from physical proximity,

colocation, and re-wiring cables in a straight line, to the installation of

181

. Levine, supra note 164 (executives from Index and PubMatic ex-

changes noting adoption of header bidding resulted in a 30-50% lift in CPM

prices); Ricardo Bilton, With Header Bidding, Publishers Are Boosting CPMs by as

much as 50 Percent, DIGIDAY (Nov. 12, 2015), https://perma.cc/SZZ9-5TBM.

182

. PubMatic Staff, How PubMatic’s Header Bidding Technology Helped an

Entertainment and News Media Company Increase CPMs By 50%, PUBMATIC (Apr.

28, 2016), https://perma.cc/343S-4JQ9 (stating that the adoption of header bidding

caused entertainment and news publisher CPMs to increase on average by 50%

and caused a daily revenue increase of 9-16X on Pubmatic’s exchange).

183

. Alex Groth, Header Bidding Is a Game Changer for Smaller Publishers,

ADMONSTERS (Nov. 3, 2016), https://perma.cc/JSF9-PMUJ (noting that publisher

and ad network Gladly states average daily revenue increased by 40% after the

first day of routing to multiple exchanges at the same time and that average rev-

enue in the long-term increased 100%).

184

. PulsePoint Enables Publishers to Bypass Waterfall Inefficiencies with

Header Bidding; Average CPM Increases 147%, BUS. WIRE (Apr. 21, 2016),

https://perma.cc/NU6A-QC5X (announcing that the adoption of header bidding

caused PulsePoint publisher CPMs to increase 147% on average).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

133

new transatlantic cables and cross-country microwave networks.

185

In ad-

vertising, when Google could no longer preferentially route ad space to

its own exchange, the tug-of-war pivoted increasingly to speed. Just as

publishers and advertisers were adopting header bidding at a rapid pace

to work around Google’s restrictions related to routing, speed, and trans-

parency, Google introduced a software development framework called

Accelerated Mobile Pages (AMP).

186

At a simple level, AMP is a frame-

work that websites can use to create fast-loading mobile web pages. In

part by limiting the types of programming codes that can be used on a

page, AMP pages load faster than they otherwise would.

187

However,

185

. For an overview, see AQUILINA ET AL., supra note 24.

186

. Google announced the AMP Project on October 7, 2015 and first

launched it early 2016. Though AMP is open source (meaning, it is developed as

a public collaboration), it is widely viewed as a Google initiative because Google

employees spearheaded its development. As a result, the news media frequently

refers to AMP simply as “Google’s AMP.” See, e.g., Sarah Sluis, Google's Acceler-

ated Mobile Pages Aim to Speed Content and Ad Load Times, ADEXCHANGER (Oct. 7,

2015), https://perma.cc/HD8K-ZA79; Mark Scott, Google’s Mobile Web Dominance

Raises Competition Eyebrows, POLITICO (June 1, 2018), https://perma.cc/TF7S-

QFYG; Danny Crichton, Answering Its Critics, Google Loosens Reins on AMP Project,

TECHCRUNCH (Sept. 18, 2018), https://perma.cc/VLF7-RRMT; A Letter About

Google AMP, AMPLETTER.ORG (Jan. 9, 2018), https://perma.cc/5XN5-3B5R; Frederic

Lardinois, Google’s AMP Project Expands Beyond News, TECHCRUNCH (Aug. 2,

2016), https://perma.cc/VLF7-RRMT; Julia Greenberg, Google's AMP Will Make the

Mobile Internet Faster Early Next Year, WIRED (Dec. 9, 2015),

https://perma.cc/4GMV-TR7Z; Scott Gilberston, How Google’s AMP Project Speeds

Up the Web—By Sandblasting HTML, ARS TECHNICA (Nov. 3, 2015),

https://perma.cc/E8EE-FRGH.

187

. The speed at which web pages load is important for consumers and

publishers and there are different ways that consumers and publishers can get

pages to load faster. For instance, consumers might pay their internet service pro-

vider more for quicker download speeds or publishers can follow best practices

in web design to optimize page load speeds. AMP pages load fast because they

limit the use of JavaScript, only permit asynchronous JavaScript execution, and

are cached. For further discussion of AMP and its restrictions, see How AMP

Works, AMP.DEV (2020), https://perma.cc/XBN7-4UU3; How AMP Pages Are

Cached, AMP.DEV (2020), https://perma.cc/SAE8-9JKQ.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

134

there are caveats to Google’s approach to speed with AMP. The AMP

framework restricts the use of JavaScript, which is precisely the code that

websites needed to use to make the client-side header bidding invention

work.

188

Google framed AMP as something that maximizes consumer welfare:

faster loading pages are better for consumers. But the question of whether

faster-and-faster page speeds maximize consumer welfare is more com-

plicated. If John is reading a news site and a page loads too slow, he might

quickly switch to another site in the hopes of finding a similar article that

loads faster.

189

The sites that keep up with John’s preference for speed

188

. Later, RTC wrappers were developed to permit publishers to route

their ad space through header bidding. However, because these RTC wrappers

impose tight restrictions on the number of entities that publishers can route to (a

max of five vendor calls overall), only permit one vendor cookie-sync, and im-

pose strict trading timeouts (of 1000 ms), the industry has continued to speak of

AMP as being simply incompatible with client-side header bidding, which was

founded on publishers being able to control both these things. See AMP Real Time

Config, GITHUB (2020), https://perma.cc/39PA-LQYK; Monetizing Your AMP Page

with Ads, AMP.DEV (2020), https://perma.cc/4J9K-CEGP (“In non-AMP pages

(traditional HTML), if you want to display ads on your page, you'd include a

snippet of JavaScript to serve ads from your ad network. For performance and

security reasons, you cannot include third-party JavaScript in AMP pages. So, to

display ads in AMP, you need to add the custom amp-ad component to your

AMP page.”); Geradin & Katsifis, supra note 158 (discussing AMP’s incompati-

bility with traditional client-side header bidding). More recently, AMP has loos-

ened restrictions around the use of JavaScript but remains incompatible with cli-

ent-side header bidding in its original form. amp-script: AMP [heart] JS, AMP.DEV

(2020), https://perma.cc/FT93-UFS6; Amp-Script, AMP.DEV (Aug. 21, 2019),

https://perma.cc/7RLC-NQSP (acknowledging that the availability of JavaScript

was “one of the most important requests from developers using AMP”).

189

. According to Google, pages that take longer than three seconds to

load lose 53% of mobile visits. However, it is unclear whether this statistic ex-

cludes bot traffic. David Kirkpatrick, Google: 53% of Mobile Users Abandon Sites

That Take Over 3 Seconds to Load, MARKETING DIVE (2020), https://perma.cc/BSL5-

6LMA. But see Jake D. Brutlag et al., User Preference and Search Engine Latency,

Joint Statistical Meeting Proceedings (2008) (demonstrating that user preference

for faster search engines kicks in at or after the three second latency mark).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

135

will have more ad space to sell, which is how page speed also correlates

with publisher welfare.

190

However, publishers also face trade-offs when it comes to speed that

ultimately impact consumer welfare. When AMP first launched, AMP

pages loaded so fast, they sometimes loaded for users before sites could

return ads at all.

191

AMP pages were also incompatible with JavaScript,

the code that publishers relied on to route their ad space to multiple ex-

changes with header bidding. If publishers make less revenue selling ad

space, they have less to re-invest into the content that consumers want to

consume. Thus, at a certain point, fast-loading pages come back to hurt

consumer welfare.

Nevertheless, Google started to condition premium treatment on

Google Search—a market where Google has 88% market share in the

U.S.—on publishers’ migration to AMP. For example, to have news arti-

cles displayed at the top of Google Search results, in the “News Carousel”

that features news articles in a visual and swipe-able format, websites

generally have to use AMP.

192

Google Search also includes a lightning

190

. Consumers now use their phones more than their computers and

publishers receive approximately 60% of their traffic from mobile use. Mason

Walker, Americans Favor Mobile Devices Over Desktops and Laptops for Getting News,

PEW RSCH. CTR. (Nov. 19, 2019), https://perma.cc/67NV-4LB3 (Pew Research sur-

vey showing roughly 6 in 10 US adults access news from a mobile device); Sluis,

supra note 186 (“The Washington Post, for one, said its audience is 60% mobile,

roughly average for most publishers.”); BRIGHTEDGE 2018 MID-YEAR MOBILE

RESEARCH ROUNDUP, BRIGHTEDGE RSCH. (2020), https://perma.cc/TPR8-SKV5

(“62% of traffic in a large sample coming from mobile and tablet search.”)

191

. Garett Sloane, Publishers Find Google’s Amp Speeds Up Pages, But Ads

Are Still Slow, DIGIDAY (Apr. 21, 2016), https://perma.cc/WEU4-PK25; see also Sa-

rah Sluis, With Ads in Place, Google Takes Training Wheels Off of AMP,

ADEXCHANGER (Feb. 18, 2016), https://perma.cc/DY36-MWJG (“Faster-loading

pages may lead to a rise in page views. But AMP doesn’t support more complex,

higher-paying ads, like interstitials, which could dampen yield.”).

192

. Search Engine Market Share United States of America, supra note 111;

Barry Schwartz, Google Reserves News Carousel for AMP Pages Only, SEARCH

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

136

bolt icon next to AMP pages in the search results, signaling to users that

those pages load fast.

193

To participate in the News Carousel, or have a

lightning bolt, a publisher has to use AMP and forgo the implementation

of header bidding using JavaScript.

With these developments, the advertising technology companies like

PreBid and Rubicon Project that developed the original header bidding

inventions started to build a modified version that complied with Google

AMP JavaScript restrictions and improved on the original version’s la-

tency challenges.

194

To speed things up and work around AMP JavaScript

constraints, these vendors moved auctions back to third-party servers,

which is commonly called “server-side” header bidding.

It was around this time, in April of 2016, that Google’s ad server re-

lented and started letting some publishers route their ad space to multi-

ple exchanges at the same time.

195

Google encouraged publishers to start

ENGINE ROUNDTABLE (Mar. 2, 2016), https://perma.cc/CM5G-KBMB (reporting

that Google announced that News Carousel is only open to publishers using

AMP); Google Developers, Structured Data: Article, GOOGLE, (Feb. 7, 2020),

https://perma.cc/62LS-SLBP (stating Google has since opened up the News Car-

ousel to non-AMP pages but only when other pages in the News Carousel are

not AMP pages); Greg Sterling, Google Adding More News Content to ‘Top Stories’

in Search Results, SEARCH ENGINE LAND (Dec. 11, 2019), https://perma.cc/QH8X-

XUNS (noting that all top pages in the News Carousel “appear to be AMP for-

matted”). But see Evaluating Page Experience for a Better Web, OFF. GOOGLE

WEBMASTER CENT. BLOG (2020), https://perma.cc/6WGT-9QV3 (announcing that

Google will soon no longer require a site to use AMP to be included at the top of

Google Search mobile stories).

193

. Lucia Moses, The Guardian Is Getting 60 Percent of Its Google Mobile

Traffic from AMP, DIGIDAY (Mar. 8, 2017), https://perma.cc/2VFY-5CY6(stating

that users increasingly recognized that the lightning bolt meant that pages would

load instantly).

194

. See generally Server Side Header Bidding Explained, AD OPS INSIDER (Jan.

25, 2017), https://perma.cc/S99F-R4EQ; Sarah Sluis, Header Bidding Goes Server-

Side: 6 Things You Should Know, ADEXCHANGER (Jan. 11, 2017),

https://perma.cc/73BS-GFB7.

195

. Google piloted the capability on April 13, 2016, beta launched it on

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

137

routing their ad space to exchanges synchronously directly from their

Google ad server. Implicitly acknowledging how its legacy routing re-

strictions had harmed market participants, Google now encouraged pub-

lishers to route their space to multiple exchanges to augment advertising

revenues by as much as 40%.

196

However, the server-side versions of header bidding re-introduce in-

formation and speed asymmetries between bidders, which can dampen

competition. Specifically, they impose tighter latency restrictions and re-

June 8, 2017, and officially launched it on April 4, 2018. Google initially called the

feature Exchange Bidding but renamed it in August 2019 to Open Bidding. See

Jonathan Bellack, Improving Yield, Speed and Control with DoubleClick for Publishers

First Look and Exchange Bidding, GOOGLE (Apr. 13, 2016), https://perma.cc/LMU4-

ENCQ; Sam Cox, Announcing Exchange Bidding Open Beta, GOOGLE AD MANAGER

BLOG (June 8, 2017), https://perma.cc/NE43-E723; Ronan Shields, Google An-

nounces Full Rollout of Exchange Bidding to Counter Header Bidding Rise, DRUM (Apr.

4, 2018), https://perma.cc/G7NW-7FZZ; Authorized Buyers Help, Publisher Inven-

tory Transitioning to a Unified First-price Auction, GOOGLE (2020),

https://perma.cc/2LRT-R2B3 (announcing renaming to Open Bidding).

196

. Shields, supra note 195 (“The rollout has taken place a little under a

year since Google first launched Exchange Bidding in open beta (after an earlier

12 month private pilot scheme) with the online behemoth claiming that trials had

seen earlier participants bolster programmatic revenues by as much as 40%.”);

Minda Smiley, OpenX Says Google's Exchange Bidding Has Led to 48% Revenue Lift

for Publishers, DRUM (Feb. 15, 2018), https://perma.cc/LRM9-NJAM; see also Jessica

Davies, Why El Pais Owner Prisa Is Shifting from Header Bidding to Server-side Bid-

ding in 9 Markets, DIGIDAY (Sept. 22, 2017), https://perma.cc/P52D-34HW (Spanish

publisher Prisa shares that “Google’s server-side option, exchange bidding in dy-

namic allocation . . . has seen decent yield lifts compared to what Google’s ad

exchange is generating”). From an antitrust perspective, given Google’s own ev-

idence of empirical price increases, it is difficult to estimate how Google might

justify its historical conduct by pointing to the conduct’s beneficial effects. For

the burden that antitrust defendants face under such scenarios, see C. Scott

Hemphill, Less Restrictive Alternatives in Antitrust Law, 116 COLUM. L. REV. 927

(2016).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

138

sult in higher cookie syncing loss rates, resulting in more bidders not be-

ing able to identity users loading pages.

197

Meaning, more bidders may

bid less for ad space and be excluded altogether due to latency.

Today, when publishers route their inventory to non-Google ex-

changes from their Google ad server, these problems persist. When it

comes to speed, Google’s ad server gives exchanges a 160 millisecond

timeout, which some people in the industry have noted is too little time

for non-Google exchanges to conclude their own auctions and return a

bid to Google’s server.

198

Increased information asymmetry between

Google and non-Google bidders exacerbates this problem. The non-

Google intermediaries can deduce the identity of users associated with

ad space even less often than they could before, further depressing their

ability to compete against DV360 and Google Ads on equal footing.

199

197

. See Lynne Johnson, Client-Side vs. Server-Side? It’s a Draw,

ADMONSTERS (July 23, 2019), https://perma.cc/89KA-STR7 (reporting on server-

side header bidding’s approach to speed and challenges with cookie matching).

Server-side auctions are also not transparent, because auctions occur inside of a

company’s servers, rather than within users’ browsers.

198

. Kean Graham, Google's Plans May Not Eliminate Header Bidding,

ADEXCHANGER (July 1, 2016), https://perma.cc/9AB2-U3JM (where an executive

of an advertising consultancy notes that “DoubleClick Ad Exchange will also

have a major advantage because of tiny timeouts for third-party ad networks,

making it more difficult for them to submit winning bids in time”). But see Google

Ad Manager, supra note 45 (stating that 160 millisecond timeouts in Google’s

server-to-server solution decreases bid exclusion).

199

. With client-side header bidding, an exchange (such as Index’s ex-

change) retrieves the ID of the user loading a page by reading the user’s ID con-

tained in the exchange’s cookies on the user’s browser. With Google’s multiple-

exchange bidding approach, competing exchanges cannot directly access the

user IDs contained on users’ browsers, and must conduct an additional cookie-

sync with Google, which increases loss rates. As the firm intermediating this pro-

cess, Google does not face the same disadvantage. See generally APPNEXUS,

HEADER BIDDING: THE NEXT EVOLUTION 10 (2017), https://perma.cc/9H9T-PEXU

(“The reason [server-side header bidding] lowers cookie match rates is that in

client-side header bidding, the header makes ad calls from the browser, where

cookies are stored, directly to programmatic demand partners. But in [server-

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

139

Google’s approach to multiple-exchange bidding restricts competi-

tion between non-Google exchanges and buying tools in more brazen

ways too. For example, when publishers choose to route their ad space

from their Google ad server to multiple exchanges at the same time,

Google’s ad server requires them to route their space through Google’s

trading venue and charges an additional intermediary fee for transacting

through a non-Google exchange, which is an additional 5-10% of the win-

ning buyer’s bid.

200

The fee would be on top of the 10-20% fee that market

side], the auction takes place away from the browser inside a third-party server,

and relies on user syncing between the third-party server and the various de-

mand partners. As a result, it’s tougher for advertisers and publishers to sync on

user identity.”). In addition, if the page the user is loading is a Google AMP page,

Google may be able to obtain additional information advantages about the user

and the content. See Scott, supra note 186 (“Other publishers gripe that AMP gives

Google access to coveted user data because almost all of AMP traffic currently

runs through the company's own servers.”).

200

. To be clear, Google’s ad server permits publishers to not do business

with Google’s exchange if they forgo routing their space to multiple exchanges

at the same time through Open Bidding. See Google Ad Manager, supra note 45

(“With Open Bidding, you can invite trusted third-party exchanges to compete

together in real-time alongside Ad Exchange in Dynamic Allocation.”); Google

Ad Manager Help, Create and Manage Yield Groups, GOOGLE (2020),

https://perma.cc/7Q7Z-K3LN (“Yield groups always include Ad Exchange, and

you can also add third-party exchanges and ad networks.”); Sarah Sluis, 3 Auc-

tions Rule Digital Advertising. Here’s a Guide to Navigating Them, ADEXCHANGER

(Nov. 20, 2019), https://perma.cc/H3HW-AYWV (“Google open bidding charges

the highest fee, a 5% fee for display and outstream inventory and a 10% fee for

video and app. That charge is on top of the exchange fee. So an exchange would

take its cut from the buyer’s bid, then Google would take its fee.”); George Slefo,

Google’s Answer to Header Bidding Runs into Headwinds, ADAGE (July 13, 2017),

https://perma.cc/9MN8-KUPL (confirming the same). In other words, if a buyer

and seller go through Google’s ad server to route and clear ad space in the Index

exchange, they must pay a fee to Index (e.g., 7% of $1,000) and an additional fee

to Google for using Index and not Google’s exchange (e.g., an additional 10%,

which would further reduce the publisher’s share to $837). When websites use

client-side header bidding to route their ad space to non-Google exchanges, pub-

lishers and advertisers have no such additional intermediary fee to absorb. See

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

140

participants already pay to exchanges.

201

By raising rivals’ costs, Google

dissuades buyers and sellers from routing trades through non-Google

venues.

202

In addition, Google’s ad server depresses competition through

a price parity rule called “Unified Pricing” that it introduced in 2019.

203

With Unified Pricing, Google’s ad server started to prohibit publishers

from routing their ad space to different intermediary buying tools and

Sluis, 3 Auctions Rule, supra note 200 (confirming that the client-side header bid-

ding vendor “Prebid doesn’t charge fees because it’s an open-source solution.”).

201

. Sarah Sluis, Explainer: More on the Widespread Fee Practice Behind the

Guardian's Lawsuit vs. Rubicon Project, ADEXCHANGER (Mar. 30, 2017),

https://perma.cc/5TFE-5L6A (stating that exchanges usually charge publishers

10-20% of ads’ clearing prices); Sluis, supra note 200 (noting that exchanges take

their cut from the buyer’s bid); Sarah Sluis, Rubicon Got Rid of Its Buy-Side Fees –

But Who Else Is Charging Them?, ADEXCHANGER (Nov. 8, 2017),

https://perma.cc/C7YF-37AB (reporting that, according to a Google spokesper-

son, publishers pay Google’s exchange a transparent revenue share that varies);

Shailin Dhar, Ad-Tech Tax — What Bracket Are You In?, MEDIUM (Apr. 7, 2016),

https://perma.cc/LWF9-J8ED (explaining generally how ad fees work).

202

. From an antitrust perspective, this can be thought of as a mechanism

to raise rivals' costs. For a discussion of this conduct and consideration under

antitrust law, see Steven C. Salop & David T. Scheffman, Raising Rivals’ Costs, 73

AM. ECON. REV. 267 (1983).

203

. Sarah Sluis, Google Switches to First-Price Auction, ADEXCHANGER (Mar.

6, 2019), https://perma.cc/8PT9-VDY7 (discussing Google plans to launch a uni-

fied pricing tool); Sluis, supra note 104 (discussing the beta launch of unified pric-

ing restrictions). There are additional ways that Google’s ad server restrictions

competition. For example, Google’s ad server shares superior information about

publishers’ auctions and pricing with competing exchanges that integrate

through Google’s ad server’s Open Bidding feature. Google easily could pass the

same information to header bidding auction vendors through an API, for exam-

ple, but chooses not to. Google markets this information advantage as a reason

to forgo header bidding and integrate through Open Bidding. Bigler, supra note

135 (“For Authorized Buyers and Open Bidders (formerly known as Exchange

Bidding buyers), we’ll provide additional information post-auction to help in-

form your bidding strategies. Buyers will receive the minimum bid price to win

after the auction closes. This feedback will help you understand when to bid

higher to win valuable inventory.”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

141

through different exchanges at different price floors.

204

The change bans

publishers from routing their ad space to, for example, Google Ads, at a

price floor of a $10 CPM, while routing the same space to The Trade Desk

at a price floor of a $9 CPM.

Historically, many publishers set higher price floors for Google Ads

because of the information advantages that Google reserves for Google

Ads, which were discussed in Part III.A. “Many set Google AdWords at

a higher floor price, since its demand often bids very low due to its supe-

rior data,” explained industry publication AdExchanger.

205

Setting lower

price floors for the non-Google intermediaries was how publishers could

manufacture competition from bidders with information disadvantages.

Publishers’ practice of setting higher price floors for the party with

information advantages—Google—was rational and output enhancing.

For example, in discussing insider trading and information advantages,

the late professor and legal scholar Victor Brudney explained: “A rational

buyer (or seller) in a market, who knows that the person with whom he

is dealing has material information about the value of the product being

exchanged which he could not lawfully acquire, will either refrain from

dealing with that transactor or demand a risk premium.”

206

In other

204

. Specifically, publishers can set price floors by advertiser and brand,

but not by “buyer,” which means not by buying tool such as Google DV360 or

Google Ads. This effectively operates as a price restraint on exchanges too. See

Google Ad Manager Help, supra note 104 (“Advertiser- and brand-specific floor

prices can be configured in unified pricing rules. However, per-buyer floor prices

are not available.”). Interestingly, Google justifies Unified Pricing rules by argu-

ing that unified price floors serve publisher interests. See Bitton & Lewis, supra

note 15, at 21.

205

. Sluis, supra note 104 (“First, publishers often set different floor prices

for different platforms. Many set Google AdWords at a higher floor price, since

its demand often bids very low due to its superior data. Under the new rules,

publishers can’t set rules at a platform level, which Google calls the ‘buyer’

level.”).

206

. See Brudney, supra note 60, at 356 ( “If the market is thought to be

systematically populated with such transactors some investors will refrain from

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

142

words, if parties cannot demand rational risk premiums when dealing

with counterparties with information advantages, they can refrain from

dealing altogether. Google’s Unified Pricing rules prohibit sellers from

asking for such risk premiums and, in the short term, let DV360 and

Google Ads buy ad space at the lower floors publishers historically set

for rivals, or alternatively, exclude rival bidders through the higher price

floors publishers historically set for Google.

From an antitrust and competition policy perspective, Google’s ap-

proach with Unified Pricing is against the grain of the trend. More and

more, these types of price parity provisions have been challenged from a

policy and antitrust perspective.

207

E-commerce retailer Amazon

dropped similar provisions after they came under scrutiny in the U.S. and

dealing altogether, and others will incur costs to avoid dealing with such trans-

actors or corruptly to overcome their unerodable informational advantages.

None of those responses is socially useful. All raise the cost of capital. And some

are simply unlawful attempts to obtain information corruptly from those who

are forbidden from disclosing it.”).

207

. See Jonathan Baker & Fiona Scott Morton, Antitrust Enforcement

Against Platform MFNs, 127 YALE L. J. 2176 (2018); Andre Boik & Kenneth S. Corts,

The Effects of Platform Most-Favored-Nation Clauses on Competition and Entry, 59 J.L.

& ECON. 105 (2016); Ariel Ezrachi, The Competitive Effects of Parity Clauses on

Online Commerce, 11 EUR. COMPETITION J. 488 (2016); Letter from Sen. Richard Blu-

menthal to Assistant Att’y Gen. Makan Delrahim (Dec. 19, 2018),

https://perma.cc/NY2C-4PBR (urging the Department of Justice to investigate the

anticompetitive effects of the price parity provisions in Amazon’s contracts with

third parties); Joe Palazzolo, Apple Ruling Heaps Doubt on 'MFN' Clauses, WALL

ST. J. (July 14, 2013), https://perma.cc/4U9Y-LZVA.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

143

globally.

208

The Department of Justice (DOJ) and U.S. courts also disfa-

vored similar price restrictions imposed by the electronic marketplaces

for airfare tickets.

209

208

. Amazon, the largest e-commerce retailer in the U.S. and globally,

nixed its price parity clauses after increased scrutiny by scholars and others. See

Catherine Shu, Amazon Reportedly Nixes Its Price Parity Requirement for Third-Party

Sellers in the U.S., TECHCRUNCH (Mar. 11, 2019), https://perma.cc/PWE9-SHFX

(reporting that Amazon is dropping its price parity provisions with third-party

sellers); Amazon to Alter Pricing Policy for Traders, BBC NEWS (Aug. 29, 2013),

https://perma.cc/F3XK-74GZ (reporting that Amazon dropped similar price par-

ity provisions in Europe after investigations by the Office of Fair Trading in the

UK and the Federal Cartel Office in Germany).

209

. The DOJ and DOT thought such clauses restricted competition be-

tween marketplaces, acted as a barrier for new entrants, resulted in effects similar

to tying, and ultimately harmed the airlines and consumers. See Computer Res-

ervations Systems (CRS) Regulations, Statements of General Policy, 67 Fed. Reg.

69365 (Nov. 15, 2002) (codified at 14 C.F.R. 255) (“We based our prohibition of

the enforcement of the systems' parity clauses on findings that those contract

provisions had the harmful effects of tying provisions—they limited competition

between the systems, and they increased the prices paid by the systems' custom-

ers.”); Computer Reservations Systems (CRS) Regulations 69 Fed. Reg. 999 (Jan-

uary 7, 2004) (codified at 14 C.F.R. 255) [hereinafter, “CRS Regulations”] (Section

E discussing the history and reasoning behind the most-favored-nation clauses

prohibition); Am. Airlines, Inc. v. Travelport Ltd., No. 4:11-CV-244-Y, 2011 U.S.

Dist. LEXIS 166373 (N.D. Tex. Nov. 21, 2011) (summarizing how regulation his-

torically and until 2004 prohibited the airfare marketplaces from imposing MFN

provisions in their contracts and how the DOT and DOJ were concerned the pro-

visions foreclosed competition between distribution channels). When the DOT

issued its final deregulation order, it warned that “clauses requiring participating

airlines to provide all fares as a condition to participation may similarly consti-

tute unfair methods of competition because they unreasonably limit each air-

line’s ability to choose how to market its services.” CRS Regulations. The Second

Circuit recently considered a similar restraint by a leading airfare marketplace in

U.S. Airways v. Sabre. U.S. Airways v. Sabre Holdings, 938 F.3d 43 (2d Cir. 2019).

In that case, U.S. Airways challenged the fact that Sabre, the leading GDS in the

U.S. with 50% market share, had a provision in its contract (called “No Dis-

counts”) that prohibited the airline from routing lower priced tickets into com-

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

144

Google’s approach is also in direct conflict with the policy approach

that lawmakers take with electronic trading in the equities market. There,

intermediaries (e.g., brokers) must act in their customers’ best interest.

210

Brokers cannot preferentially route order flow to their owned and oper-

ated exchange unless doing so best serves the client’s interest (e.g. best

price).

211

Because Google’s trading intermediaries could negotiate out of

the duties they owe to publishers and advertisers and no equivalent trad-

ing rule exists, Google can adopt a price parity rule that ultimately harms

buyers and sellers.

4. Google Search “Speed Update” Further Restricts Trading

Through Non-Google Venues

Google soon gave publishers a second reason to adopt AMP and

forgo header bidding. In 2018, Google Search released a “Speed Update”

and started to rank mobile pages according to how fast they load.

212

Spe-

cifically, Search started to push pages that were “slow” down in users’

peting marketplaces. Another provision called “No Better Benefits” required air-

lines to provide all available inventory to Sabre’s GDS. The district court ruled

that Sabre’s restrictions were illegal vertical restraints that violated Section 1 of

the Sherman Act. However, the Second Circuit reversed and remanded the case

in light of the Supreme Court’s recent holding in Ohio v. Amex, 138 S. Ct. 2274,

(2018), which rendered the district court’s jury instructions around market defi-

nition erroneous.

210

. See discussion in supra note 54.

211

. Id.

212

. Zhiheng Wang & Doantam Phan, Using Page Speed in Mobile Search

Ranking, GOOGLE WEBMASTER CENTRAL BLOG (Jan. 17, 2018),

https://perma.cc/F972-ZDJV (announcing full roll-out of “Speed Update”); Addy

Osmani & Ilya Grigorik, Speed Is Now a Landing Page Factor for Google Search and

Ads, GOOGLE (Sept. 23, 2019), https://perma.cc/7AQT-QJ6H. Note, Google started

using speed as a ranking factor for desktop pages in 2010, but “relevance” for

those pages remained the “primary” ranking signal. Additionally, on May 28,

2020, Google announced upcoming changes to how it ranks Search pages. With

these changes, Google will start to consider a page’s “Core Web Vitals,” which

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

145

search results. Rather than providing websites with an objective speed

requirement or tools to determine speed compliance,

213

Google steered

websites to a speed compliant solution: “if you're looking for a fast-by-

default framework for your pages, take a look at AMP.”

214

Publishers that

did not want to lose their Google Search traffic—which for news publish-

ers can account for 40-50% of incoming traffic—had to comply with this

Speed Update.

215

For its part, Google maintains that it does not have an AMP require-

ment, only a speed requirement.

216

But many publishers have complained

includes speed and other factors, and, will no longer require a site to use AMP to

be included at the top of Google Search mobile stories. Sowmya Subramanian,

Evaluating Page Experience for a Better Web, WEBMASTER CENTRAL BLOG (May 28,

2020), https://perma.cc/XMF5-ATFH.

213

. See Osmani & Grigorik, supra note 212; Wang & Phan, supra note 212

(where Google acknowledges “there is no tool that directly indicates whether a

page is affected by this new ranking factor”). Although Google does not provide

a direct tool for publishers to determine speed compliance, Google does offer

tools such as the Lighthouse Report, the Speed Scorecard and Impact Calculator,

the Chrome User Experience Report, and PageSpeed Insights tool to generally

advise about speed performance. See Mark McGonigle, Slow Down and Think

About Google’s Speed Update, DRUM (July 30, 2018), https://perma.cc/YU89-6KJM.

See also A Letter about Google AMP, supra note 186 (where more than 600 signees

pen an open letter to Google urging the company to adopt a neutral, objective

speed criterion).

214

. Osmani & Grigorik, supra note 212.

215

. Lucia Moses, How Google Is Using Its Search Clout to Steer Publishers to

Use AMP, DIGIDAY (Feb. 15, 2018), https://perma.cc/67MB-EJWL (stating that

Google makes up about 40% of Chartbeat publishers’ referral traffic); Kelsey Ar-

endt, Publishers, This Is Where Your Traffic Is Going to Come from in 2019, WHAT’S

NEW IN PUBLISHING (Feb. 12, 2019), https://perma.cc/3PFS-4MXE (stating that

Parse.ly powered sites received over 50% of their external traffic from Google

Search in 2018). For a discussion on the correlation between publisher AMP use

and traffic loss, see Scott, supra note 186.

216

. Understand How Amp Looks in Search Results, GOOGLE SEARCH (July 14,

2020), https://perma.cc/4GAL-6FTR ( “While AMP itself isn't a ranking fac-

tor, speed is a ranking factor for Google Search. Google Search applies the same

standard to all pages, regardless of the technology used to build the page.”).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

146

that the only way to comply with the Speed Update is by using AMP,

leading many to effectively equate the Speed Update with an AMP re-

quirement, despite Google’s claims to the contrary.

217

Overall, AMP guar-

antees publishers higher positions in Google Search, so they feel com-

pelled to use it given the correlation between the number of visitors that

Search directs to their website and the quantity of ad space they have to

sell.

218

Reflecting on this catch-22, industry publication Digiday noted:

“[i]n theory, adoption of AMP is voluntary. In reality, publishers that do

not want to see their search traffic evaporate have little choice.”

219

Google’s particular approach with AMP and speed may explain why

AMP pages would indeed load fastest for Google. In addition to limiting

the deployment of JavaScript code on webpages, Google stores AMP

pages on Google’s servers.

220

In other words, AMP pages are colocated

217

. Over a half dozen prominent U.S. and European publishers shared

this sentiment with Politico. Scott, supra note 186 (where Barry Adams from me-

dia market consultancy Polemic Digital shared that their publisher clients see a

200-300% lift when using AMP and that “[p]ublishers are held hostage by

Google,” and where some publishers also shared that their pages are penalized

in Google Search results if they do not use AMP); A Letter About Google AMP,

supra note 186 (open letter to Google by 600+ signees complaining about Google

Search’s preferential treatment of AMP pages).

218

. Scott, supra note 186.

219

. Moses, supra note 215. See also Ross Benes, Publishers Find Google AMP

Loads Too Fast for Ad Views, DIGIDAY (Oct. 9, 2017), https://perma.cc/U6Y6-7DK8

(industry executive stating that “[t]he whole reason that publishers are consider-

ing AMP is that Google gives AMP pages prioritization in search,” and “[o]ne

person’s page speed is another person’s monetization problem”). But see also

Gabe Bender, Ads and AMP: Year in Review and Looking Ahead, AMP (Feb. 14,

2018), https://perma.cc/LZG2-GU56 (stating that publishers have tripled the

amount of money they’re making from AMP pages in the past year and sped up

the load time of ads).

220

. Scott, supra note 186 (“The technology allows publishers’ mobile sites

to be stored in Google’s own servers and preloaded on individuals’ smartphones,

significantly reducing lag time when people are surfing the mobile web on

patchy cellphone networks.”); A Letter about Google AMP, supra note 186 (“Search

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

147

with Google. Colocation makes it physically impossible for another speed

framework to match Google’s AMP speeds.

With the Google Search Speed Update pigeonholing websites into

further adopting AMP, attention increasingly turned to AMP repercus-

sions on page monetization. Some publishers that adopted AMP reported

they either did not make more money per page or made significantly

less.

221

Google responded by releasing data rebutting these concerns. It

pointed to internal data showing that AMP pages were generating three

times more revenue per day.

222

However, what some people might not

have noticed was that this was not three times more revenue overall. Ra-

ther, it was three times more revenue only from Google’s proprietary ex-

change and buying tools, which are also colocated with Google.

engines are in a powerful position to wield influence to solve this problem. How-

ever, Google has chosen to create a premium position at the top of their search

results (for articles) and a “lightning” icon (for all types of content), which are

only accessible to publishers that use a Google-controlled technology, served by

Google from their infrastructure, on a Google URL, and placed within a Google

controlled user experience.”); Martin Schierle, Measuring AMP Performance, AMP

(Jan. 17, 2018), https://perma.cc/NQB2-VHCK (study concluding that sites deliv-

ered from AMP's cache deliver fastest).

221

. Greg Sterling, Report: AMP Causing Monetization Frustration Among

Some News Publishers, SEARCH ENGINE LAND (Oct. 28, 2016),

https://perma.cc/4ZRL-JSUH (reporting some publishers experience a 50% de-

crease in revenue per page with mobile AMP pages); Johnson, supra note 197

(stating that server side auctions result in less revenue per thousand impres-

sions); Jack Marshall, Google AMP Gets Mixed Reviews from Publishers, WALL ST. J.

(Oct. 28, 2016), https://perma.cc/5QW9-GEF7 (The Washington Post reporting it

generates the same revenue from AMP mobile pages as from non-AMP mobile

pages; CNN reporting it also monetizes AMP and non-AMP pages at the same

rate; multiple publishers reporting that AMP pages monetize at half the rate of

non-AMP pages). But see Sluis, supra note 194 (stating that server-side auctions

result in higher bid density and higher yield).

222

. Ads and Amp, supra note 219 (citing Internal Data, GOOGLE, Feb. 2018)

(“In the past year alone, we’ve seen publishers on AMP reach several milestones:

they’re generating up to $6 million per week from ads using Google AdSense

and Doubleclick Ad Exchange, earning 3X more revenue per day, and leveraging

over 100 ad tech platforms supporting the open source AMP Project.”).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

148

More and more, the competition narratives in the ad market echo

those happening in parallel in the electronically traded market for equi-

ties. Take, for example, the NYSE’s recent installation of new microwave

antennas on the rooftop of its data center, which increase transmission

speeds between traders and the NYSE by two-millionths of a second.

223

The NYSE had announced plans in 2019 to give its affiliated network pro-

vider SFTI exclusive access to this faster speed lane,

thereby requiring fi-

nancial traders that do not want to be excluded due to latency to go

through SFTI.

224

The NYSE plans prompted complaints from financial

traders, alleging that requiring traders to go through SFTI, and SFTI only,

to access a particular speed would be anticompetitive and a violation of

U.S. antitrust laws.

225

Another thing to note about the evolution of the securities market is

how some stock exchanges have started installing “speed bumps” to in-

tentionally slow down trading, to neutralize speed races.

It was Wall

223

. Alexander Osipovich, NYSE Antennas Spark High-Speed Trader Back-

lash, WALL ST. J. (Aug. 8, 2019), https://perma.cc/FAK8-XCT5.

224

. Id.

225

. See e.g., Letter from Thomas M. Merritt, Deputy Gen. Couns., Virtu

Financial, to Brett Redfearn, Dir., Div. Trading and Mkts., SEC (June 25, 2019),

https://perma.cc/BN6G-EEBY ( “In the same manner that the NYSE would never

have been permitted to sell positions on the floor of the exchange on Wall Street

within earshot of the specialists to the detriment of the rest of the members for a

premium price so too should it be impermissible for the NYSE to do the same

with rooftop access to the new exchange ‘floor.’”); Sabri Ben-Achour, Why Are the

NYSE and Private Companies Fighting over Two-Millionths of a Second?,

MARKETPLACE (Oct. 7, 2019), https://perma.cc/6DDZ-2VKD; Self-Regulatory Or-

ganizations; New York Stock Exch. LLC, Nyse Am. LLC, Nyse Arca, Inc., Nyse

Chicago, Inc., & Nyse Nat'l, Inc.; Notice of Filings of Partial Amendment No. 3 &

Order Granting Accelerated Approval to Proposed Rule Changes, Each As Mod-

ified by Partial Amendment No. 3, to Establish A Wireless Fee Schedule Setting

Forth Available Wireless Bandwidth Connections & Wireless Mkt. Data Connec-

tions, https://perma.cc/3XHN-3ZVM announcing SEC approval of NYSE plans

on the condition that the NYSE would not have a speed advantage over wireless

data providers). Note, NASDAQ provides clients with access to faster data (the

FGPA offering) for a cost of $25,000 per month.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

149

Street veteran Brad Kutsayama who spearheaded such bumps on the

public stock exchange IEX, which the SEC approved in 2016.

226

Some ac-

ademic studies have suggested that slowing things down improves mar-

ket efficiency. For example, one study by an SEC economist studied how

slowing things down can decrease trading costs for individual inves-

tors.

227

Another paper released in January of 2020 by the U.K. financial

regulator, the Financial Conduct Authority, concluded that eliminating

the ultra-fast trading strategies used by high-frequency traders could

save investors $4.8 billion per year.

228

In advertising, however, speed nar-

ratives are beholden to Google’s power in the search market.

C. Inside Information Abuses

1. Google’s Ad Server Shares Information About Competitors’

Trading Activity with Google’s Exchange and Buying Tools,

Permitting Them to Trade Ahead of Orders

With AMP and the Search Speed Update, Google’s stated priority is

how fast webpages load for consumers. But when websites route their ad

space to exchanges through header bidding rather than through Google’s

sell-side, the only way that a website can route its inventory to Google’s

226

. Dave Michaels, IEX Gains SEC Approval for New Stock Exchange, WALL

ST. J. (June 17, 2016), https://perma.cc/TNT9-XQ4A (announcing SEC approval of

the IEX national stock exchange that slows down trading by 350 millionths of a

second).

227

. Edwin Hu, Intentional Access Delays, Market Quality, and Price Discov-

ery: Evidence from IEX Becoming an Exchange, (Div. Econ. & Risk Analysis Working

Paper Series, 2018) (finding that trading delays on securities exchanges reduced

trading costs for consumers for some stocks).

228

. AQUILINA ET AL., supra note 24; see also Jonathan Macey & Maureen

O’Hara, Regulating Exchanges and Alternative Trading Systems: A Law and Econom-

ics Perspective, 28 J. LEGAL STUD. 17 (1999) (discussing fair competition and how

unequal access to data can weaken information efficiencies and lead to larger

intermediary spreads and noting that vertical economies of scale may provide a

counterbalance to these arguments).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

150

exchange is by sequentially routing it to Google’s exchange after the space

has cleared in another in header bidding.

229

The sequence is necessary

because Google’s trading venue refuses to return bids through header

bidding.

230

The two-step setup ultimately adds latency to consumer page

loads, reserving the benefits of speed to those that stay within Google’s

Walled Garden.

For many years, the two-step setup and latency provided Google

with a competitive advantage: it permitted Google to use information

about the trading activity of rival exchanges to inform its own trading

activity in the market.

231

If a publisher like The Register routed its ad space

through header bidding, the winning bid belonging to the non-Google

exchange was passed into the publisher’s ad server—usually, Google’s.

From 2015 through 2019, Google’s ad server then passed information

229

. See, e.g., What is Prebid.js?, PREBID (2020), https://perma.cc/4DEY-VZJ6

(explaining how PreBid auctions occur before publishers’ ad server is called and

stating that the ad server starts after header bidding concludes). When publishers

do sequentially route to Google’s ad server (Google Ad Manager, or GAM), they

can then route the impression either to Google’s exchange (or a non-Google ex-

change “line item” that the publisher has set up in their ad server) or to Google’s

multiple-exchange solution Open Bidding. See How Open Bidding Works, GOOGLE

AD MANAGER HELP (2020), https://perma.cc/FEU7-AZTW. Note, this sequence

also occurs when a publisher uses a server-side header bidding vendor such as

PreBid or Amazon’s Transparent Ad Marketplace (TAM).

230

. Sluis, supra note 200 (“The Google exchange formerly known as AdX

doesn’t participate in Prebid or TAM, leaving a gap which prevents publishers

from shifting their entire auction to the header.”).

231

. Specifically, this occurred as part of Google’s “Dynamic Allocation”

setup and may have ended with Google’s unified pricing changes introduced in

the fall of 2019. People in the industry routinely discussed this Google infor-

mation advantage. See Garett Sloane, WTF is Dynamic Allocation?, DIGIDAY (Apr.

14, 2016), https://perma.cc/LWM2-F2ZZ (where chief revenue officer of media

publisher states “Google had an informational advantage to buying the best im-

pressions, and the informational advantage came from the fact that they own the

ad server”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

151

about that winning bid (e.g., $10.48) to Google’s exchange.

232

Google’s ex-

change then passed that the bid to its own bidders, including the largest

bidder by trading volume, Google Ads.

233

At the end of that circuitous

information sharing, Google’s ad server permitted Google Ads, DV360,

or another bidder in Google’s exchange to trade ahead of rivals’ orders

by paying one penny more.

234

In the equities market, the intermediary broker dealers that trade on

behalf of third parties, as well as on behalf of themselves, and even run a

trading venue, must manage conflicts of interest and use ethical walls to

prevent sensitive trading data from flowing from one business division

to another. When a broker uses information about a customer’s trading

activity to race in front of orders and trade for its own benefit, it is called

front running or trading ahead and is prohibited (and even criminal).

235

In the unregulated online advertising market in Silicon Valley, a world

232

. Id.

233

. Specifically, the bid from header bidding became the reserve price

(i.e., the floor price) that a bidder in Google’s exchange could displace. See gener-

ally Sarah Sluis, Google’s First-Price Auction Switch Is Making Header Bidding Part-

ners Win More, ADEXCHANGER (Sept. 5, 2019), https://perma.cc/57UF-WVMW (

“In the previous second-price auction model, Google reserved a ‘last look’ ad-

vantage that allowed it to bid last after everyone had already submitted their

bids. It could win impressions for a penny more than the highest bidder, allowing

it to cherry-pick valuable users and clear more impressions than its competi-

tors.”).

234

. Id.

235

. See RENA MILLER & GARY SHORTER, CONG. RSCH. SERV., R44443, HIGH

FREQUENCY TRADING: OVERVIEW OF RECENT DEVELOPMENTS (2016); Edward B.

Rock, Foxes and Hen Houses? Personal Trading by Mutual Fund Managers, 73 WASH.

U. L.Q. 1601, at 1607 (1995) (noting that front running is usually handled through

stock exchange and self-regulatory organization rules but that it could also be

construed as a violation of Section 10(b) and Rule 10b-5). Note, front running

prohibits dealers from trading ahead of client orders handled by the broker divi-

sion. In advertising, by refusing to deal directly with third-party exchanges and

intermediaries and forcing publishers instead to go through a two-step routing

process, Google technically avoided trading ahead of a third party it had a direct

relationship with.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

152

apart from Wall Street, industry insiders nicknamed this somewhat sim-

ilar setup Google’s “last look.”

236

Last look helped Google’s exchange and buying tools to further con-

solidate their respective market shares. Publishers may have discovered

header bidding to get around Google’s routing restrictions and to route

their ad space to multiple exchanges at the same time, but the Dou-

bleClick ad server, with its overwhelming market share, still set the end-

rules of the game. With header bidding, non-Google exchanges could

compete against each other all day long for publisher’s inventory only to

have Google’s ad server let Google’s exchange jump ahead of orders and

displace trades by a penny.

It wasn’t just competition that last look distorted. There were also

concerns around implications to market efficiency, ultimately measured

by trading costs. Remember, advertising auctions are normally “blind”—

buying tools return bids on behalf of advertisers without knowing what

others are simultaneously returning as their bids. With last look, one

trading division at Google (i.e., the ad server) let the other trading divi-

sions (i.e., the exchange and buying tools) see information about rivals’

bids (i.e., their winning bid). That permitted a buyer like Google Ads to

know what the bidders with the user ID information disadvantage would

pay for an ad (e.g., $10.48, not $14, CPM) and use that information to bid

lower than what it otherwise would have bid for the same space.

237

discussed above, it is difficult to say whether lower prices paid by Google

Ads benefitted advertisers on the other ends of these trades or whether

they instead permitted Google as the intermediary to keep a wider

spread.

236

. See Sluis, supra note 233.

237

. Most publishers use Google’s ad server and Google’s ad server shares

superior information about users’ identity with Google Ads and DV360 than it

shares with others. See supra Part III.A. Google’s ad server also shares with ad-

vertisers bidding through Google superior information about publisher price

floors. Sluis, supra note 200 (discussing how Google’s ad server does not share

floors with exchanges outside of Google’s Open Bidding integration).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

153

According to some discussion of the topic, Google’s ad server

stopped passing the “price to beat” to Google’s exchange in late 2019.

238

However, monitoring whether this conduct has in fact stopped is more

difficult. Around the same time that Google stopped permitting itself to

view the “price to beat,” Google started rounding down the timestamps

of bids in the consolidated auction reports it shares back with publishers.

The reports previously disclosed the time in microseconds (i.e.,

23:59:59.998877) that each exchange submitted a bid for a publisher’s

space. Now, Google would start to round down those timestamps from

microseconds to the nearest hour (i.e., 23:00:00.00000).

239

Without precise

timestamps, it becomes difficult to monitor if the ad server continues to

let Google’s exchange trade ahead of rival exchange bids. For instance,

the ad server might receive Exchange B’s bid of $10.48 first, pass infor-

mation about that bid to Google’s exchange, and let Google’s exchange

return a slightly higher bid of $10.49 a few milliseconds later. Without

precise timestamps, publishers would not know.

238

. See Sluis, supra note 200 (“[W]hen Google made its changes to unified

pricing and switched to a first-price auction this fall, it removed one distinct ad-

vantage in open bidding, known as ‘last look.’ Until a couple of months ago, open

bidding used the results of the Prebid and TAM auctions as the “price to beat”

for all the open-bidding buyers. That extra information meant that open-bidding

buyers could outbid Prebid and TAM winning bids.”).

239

. Google Ad Manager Help, Data Transfer Fields, GOOGLE (2020), supra

note 90; Ad Manager Data Transfer Reports, GOOGLE AD MANAGER HELP (2020)

https://perma.cc/8QSV-YBQF (describing the various types of reports Google’s

ad server shares with market participants); Google Ad Manager Help, Bids Data,

supra note 135 (discussing the Bid Data file NetworkBackfillBids that contains

information about bids coming in from Google’s exchange and Google Open Bid-

ding auctions; stating that Google truncates timestamps of bids to the nearest

hour); Damien Geradin & Dimitrios Katsifis, Trust Me, I’m Fair: Analysing Google’s

Latest Practices in Ad Tech from the Perspective of EU Competition Law, 16 EUR.

COMPETITION J. 11 (2020), (discussing how sellers used timestamps of bids field

called “TimeUse2” and field called “KeyPart” to tie information about bids from

Google auctions to information about bids from header bidding auctions in the

NetworkCodeServes/NetworkBackfillCodeServes file but can no longer do so be-

cause Google rounded timestamps).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

154

In the securities market, we require intermediaries to disclose trading

timestamps in milliseconds or a finer increment to help to protect against

these types of market abuses. In fact, the SEC has prosecuted some trad-

ers for redacting timestamps in order to conceal the fact that they were

trading ahead of orders. For example, in 2004, the SEC and the National

Association of Securities Dealers (predecessor to the Financial Industry

Regulatory Authority) charged the Knight Trading Group for trading

ahead of client orders and recording inaccurate trade execution times to

evade client oversight.

240

In advertising, weak competition permits a

powerful market intermediary like Google to negotiate out of timestamp

disclosures.

2. Google Amends Terms and Conditions to Breach “Ethical Walls”

Google misuses the material nonpublic information belonging to

third-party market participants in another way: by breaching ethical

walls and trading on their sensitive non-public data. Publishers protect

their financial interest in the online content they produce by limiting

third parties’ access to their readership data. For instance, say user

1Q2W3E reads The Journal’s “Heard on The Street” column about investing,

which permits that publisher to solicit high bids for its ad space from ad-

vertisers like Goldman Sachs and Barclays. If another site, say user

1Q2W3E’s email provider, knows that user 1Q2W3E reads The Journal’s

“Heard on the Street” column, the email provider can use that information

to solicit bids from Goldman Sachs and Barclays too, even though the

240

. In re Knight Securities L.P. Respondent, SEC Admin. Release No.

50867, 2004 WL 2913488 (Dec. 16, 2004) https://perma.cc/4JTN-ET2Q (discussing

how misreporting timing of trades permitted Knight to trade ahead of orders and

defraud thousands of customers); Ann Davis, How Knight Clients Failed to Recog-

nize Questionable Trades, WALL ST. J. (Dec. 17, 2004), https://perma.cc/2PHK-S75U;

Harvey L. Pitt, Conflict of Interest Lessons from Financial Services, COMPLIANCE

WEEK (Feb. 21, 2005), https://perma.cc/2HKL-GJJJ (reflecting on how the Knight

problem was a conflicts of interest problem).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

155

email provider produces no content and only knows that user 1Q2W3E

buys lots of widgets from Amazon. Simply put, by appropriating a com-

peting ad seller’s readership data, the email provider can increase the

supply of space available to users known to be interested in investing,

depressing clearing prices for that type of advertising and hurting the

content creator’s ability to monetize its original work.

Historically, when competition in the online advertising market

worked, Google safeguarded such sensitive business data behind effec-

tive ethical walls. Google became aware of such data in the course of li-

censing its buy-side and sell-side ad server to market participants. Recall,

in 2008, Google explained that Google could not “do anything” with that

data because ownership vested in publishers and advertisers.

241

As a re-

sult, Google had a duty to safeguard that data and could not use it for its

own financial gain.

However, as Google’s market power grew, Google tore down these

data separations. Shortly after acquiring DoubleClick, Google started to

restrict publishers’ and advertisers’ access to DoubleClick user IDs.

242

Then, in 2012, Google amended its terms and conditions to obtain per-

mission to merge data from DoubleClick with data from other Google

business divisions, including Google’s exchange division and Search and

YouTube.

243

In 2016, Google amended its policies again, this time to com-

bine DoubleClick data with data Google separately has about consumers’

241

. The Google-DoubleClick Merger, supra note 82.

242

. See AdExchanger Q&A, supra note 89.

243

. For a summary of Google’s 2012 privacy change, see Google Merges

Privacy Policies and Data Across Services, PRIVACY INT’L (Jan. 25, 2012),

https://perma.cc/8B6J-8VFT, which states, “In 2012, Google announced it would

condense 70 different privacy policies into a single one that would allow the com-

pany to merge the data collected across all its services, including Maps, Search,

Android, Books, Chrome, Wallet, Gmail, and the advertising service provided by

its DoubleClick subsidiary into a single database.” Impressively, the FTC Com-

missioner and public interest groups the Electronic Privacy Information Center

(EPIC), Center for Digital Democracy, and U.S. Public Interest Research Group

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

156

identity.

244

In 2018, Google started obtaining access to users’ readership

information on third-party sites and apps directly from the Chrome

browser, circumventing the need to negotiate for these rights with third

parties (e.g., publishers and advertisers).

245

Google also turns those information advantages into financial gain.

In addition to obtaining access to such data through data set mergers and

through Chrome, Google obtains permission to use that information to

buy and sell in the advertising market for its own interests. With the 2012

raised concerns in 2007 around approving a Google-DoubleClick merger without

imposing restrictions on Google’s ability to merge data sets. See Complaint and

Request for Injunction, Request for Investigation and for Other Relief, Second

Filing of Supplemental Materials in Support of Pending Complaint and Request

for Injunction, Request for Investigation and for Other Relief, and Dissenting

Statement in Google and DoubleClick, supra note 80. EPIC also challenged

Google’s 2012 plan to merge datasets by filing a temporary restraining order and

preliminary injunction compelling the FTC to force Google to abide by an Octo-

ber 13, 2011 FTC Consent Order. The 2011 Consent Order required Google to

establish new privacy safeguards for users and to subject itself to ongoing pri-

vacy audits. The federal court ultimately dismissed EPIC’s complaint on the

ground that the FTC has discretion over whether to enforce its Consent Orders

and that the exercise of such discretion is not subject to judicial review. See EPIC

v. FTC (Enforcement of the Google Consent Order), ELEC. PRIVACY INFO CTR.,

https://perma.cc/N2BE-T3S6.

244

. Angwin, supra note 105; Google Privacy Policy, GOOGLE (Mar. 31, 2020),

https://perma.cc/MS8P-WWBG (explaining to consumers that information about

their “visits to sites and apps that partner with Google,” including from sites that

use Google advertising services, may be combined with other data Google has

about the consumer “in order to improve Google’s services and the ads delivered

by Google”). Google’s merger of datasets also poses negative implications to con-

sumer privacy, which was a concern also raised by public interest groups and

FTC Commissioner Pamela Jones Harbour at the time of the Google-DoubleClick

merger. See Complaint and Request for Injunction, Request for Investigation and

for Other Relief, Second Filing of Supplemental Materials in Support of Pending

Complaint and Request for Injunction, Request for Investigation and for Other

Relief, and Dissenting Statement in Google and DoubleClick, supra note 80 (ex-

plaining that when it comes to Google and DoubleClick potentially merging da-

tasets, privacy concerns are “the other side of the coin”).

245

. Green, supra note 114 and accompanying comments.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

157

change, privacy commissioners in Europe lamented that the change per-

mits Google to “combine almost any data from any services for any pur-

poses.”

246

With the 2016 change, Google obtained specific permission to

use the data “to improve … ads delivered by Google.”

247

Today, via

Chrome, Google obtains permission to use third party readership data to

sell advertising, including advertising on its own properties Search,

Gmail, and YouTube.

248

Since 2012, Google’s decisions to merge data across internal business

divisions have primarily gained attention as harming consumer pri-

vacy—by combining the consumer records of one division with the con-

sumer records of another, Google develops deeper and deeper profiles of

consumers’ online behavior.

249

However, these data mergers also distort

competition outcomes in advertising markets. Google merges data sets in

ways that conflict with the interests of market participants.

250

Google can

use third party readership data to disproportionately sell more ads on

Google properties.

251

In financial markets, this progression cannot hap-

pen, in part because intermediaries cannot negotiate out their duty to act

in their customers’ best interest.

246

. Letter from Article 29 Data Prot. Working Party to Larry Page, CEO,

Google (Oct. 16, 2012), https://perma.cc/4LQJ-3D45.

247

. Google Privacy Policy, supra note 244.

248

. Green, supra note 114 and accompanying comments. Note, with the

2012 and 2016 changes, it is unclear whether Google terms with publishers and

advertisers separately restrict Google’s use of this data. However, with the

Chrome changes, it does not appear that any counterparty relationship with pub-

lishers and advertisers could restrict Google’s use of this parallel data set.

249

. See, e.g., Johnny Ryan, Failure to Enforce the GDPR Enables Google’s Mo-

nopoly, BRAVE (Feb. 18, 2020), https://perma.cc/3VRB-P7BZ (describing Google’s

merger of privacy policies as an “internal data free-for-all”).

250

. Specifically, the terms appear to permit Google to use information

belonging to third-party buyers and sellers obtained by the ad server division

and Chrome to buy and sell in the marketplace and on exchanges.

251

. See discussion supra note 154.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

158

The reasons that we prohibit such conduct in financial markets also

exist in advertising markets: Google gained access to confidential busi-

ness data by rendering unrelated business services, the use of such data

provides Google with an unmatched information advantage when buy-

ing and selling, and letting Google trade on those information ad-

vantages may dissuade firms from participating in online advertising

markets at all.

252

In other words, the practice is not merely unfair, it re-

flects a socially undesirable allocation of property rights in information

that can result in long-term output declines. The publisher was the party

that risked capital, hired journalists, and produced content to know that

user 1Q2W3E is interested in investing. The property right in that piece

of information should be allocated to the party that bore the costs of pro-

ducing it (i.e., the website), rather than the trading intermediaries that

websites use (i.e., Google’s sell-side ad server, Google’s exchange,

Google’s buy-side DSP, or even Chrome).

IV.POLICY CONSIDERATIONS

A. Advertising Exchanges Should Provide Fair Access to Information

and Speed

Outside of antitrust enforcement, lawmakers might lean on the prin-

ciples of financial regulation to develop a framework to protect competi-

tion in advertising markets. In 1934, Congress developed a competition

framework for the securities markets, viewing them as an “important na-

tional asset” that should be “preserved and strengthened.”

253

As this Part

explores, the electronic markets for sports tickets, theatre tickets, airfares,

252

. My argument is that because Google does not actually produce con-

tent in advertising markets, such information appropriation conduct may lead to

overall output declines, which is a traditional welfare concern. For a closely re-

lated conversation around how information misappropriation by large digital

firms may deter entry and chill innovation, see Khan, supra note 20, at 1008-15.

253

. Securities Exchange Act, 15 U.S.C. § 78k-1(a)(1)(A) (2012) (“The secu-

rities markets are an important national asset which must be preserved and

strengthened.”).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

159

and cryptocurrencies have been sufficiently important to warrant some

kind of intervention.

254

What about advertising? It is a primary driver of

the business of journalism and news, a cornerstone of democracy, and

surely worthy of equal scrutiny.

One basic rule that lawmakers ought to borrow from the equities

market and apply in advertising is a requirement that exchanges provide

all bidders with fair access to both data and speed. Google’s exchange

should share user identity information and speed with intermediaries in

a non-discriminatory manner. It could even colocate with bidders in neu-

tral colocation facilities, as other exchanges and intermediaries already

do.

Lawmakers have applied the same principle to address parallel com-

petition problems in other electronically traded markets. Exchange access

problems related to data and speed came up in the much smaller elec-

tronic trading market for event tickets. There, differential access to speed

and information was distorting competition between those buying and

selling on the new, centralized electronic ticket marketplaces.

255

For ex-

ample, buying at “lightning speed” allowed some computerized ticket

brokers to buy 1000-plus tickets per minute and over 90% of desirable

tickets to major concerts from trading venues like Ticketmaster and Stub-

hub, before consumers had enough time to check-out online.

256

The abil-

ity to exclude competition from consumers allowed intermediary brokers

to buy up tickets on initial release then resell them back to consumers at

high mark-ups—49% higher on average, but sometimes over 1000%, ac-

cording to one investigation.

257

254

. See discussion infra notes 259-61 (event tickets), 262-66 (cryptocurren-

cies), and 288-90 (airfares).

255

. Schneiderman, supra note 26.

256

. Id. at 8, 18-19 (detailing that one high-speed buyer purchased 1,012

tickets to a 2015 U2 concert in Madison Square Garden in under one minute); see

also Tod Marks, Why Ticket Prices Are Going Through the Roof, CONSUMER REPS.

(June 30, 2016), https://perma.cc/C25Y-QBS7.

257

. Schneiderman, supra note 26, at 4.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

160

To address these speed and data competition problems in the elec-

tronically traded ticket market, lawmakers in the U.S. and globally

passed legislation. Thirteen states, including New York, California, Penn-

sylvania, and Florida, passed legislation that restricts ticket brokers and

other buyers from using high-speed bot software on the centralized ticket

marketplaces to distort competition between those buying on the same

venue.

258

Congress in 2016 then extended these state restrictions to ticket

brokers nationally with the Better Online Ticket Sales Act.

259

The UK,

Australia, and Canada also passed legislation restricting speed practices

to safeguard access to ticket marketplaces.

260

Across jurisdictions, pro-

tecting fair access to the trading venues was the driving concern.

261

Similarly, lawmakers are now monitoring emerging cryptocurrency

markets for fair access problems that stem from the role that data and

speed play with electronic trading.

262

In 2018, the NYAG conducted an

258

. Robert McFadden, The BOTS Act: A Small Step for Fankind When a Giant

Leap Is Needed, 55 WASHBURN L.J. 427 (2016).

259

. Better Online Ticket Sales Act, 15 U.S.C. § 45c (2016) [hereinafter

“BOTS Act”]. The New York State legislature later made its law stronger by in-

creasing civil fines and imposing criminal penalties for violations of the New

York state anti-bot law. Press Release, Andrew Cuomo, Governor, N.Y., Gover-

nor Cuomo Signs Legislation Combating Unfair Ticket Purchasing and Reselling

Practices (Nov. 28, 2016), https://perma.cc/ATL9-MDST.

260

. Digital Economy Bill 2017, Dep’t. for Culture, Media, and Sport Supp.

Memo (UK) (prohibiting the use of bots to circumvent ticket marketplace rules);

Ticket Sales Act, S.O. 2017, c. 33, Sched. 3 (Can.) (prohibiting the use of bot soft-

ware to circumvent ticket marketplace access rules); Fair Trading Amendment

(Ticket Scalping and Gift Cards) Act 2017, No. 52/2017 (NSW) (amending the Fair

Trade Act of 1987 to include regulation of ticket scalping and gift card expiry

dates).

261

. See, e.g., BOTS Act, supra note 259 (stating aim of legislation is to “en-

sure equitable consumer access to tickets”).

262

. In the U.S., the SEC, the Commodity Futures Trading Commission

(CFTC), and the New York State Department of Financial Services share some

oversight over the cryptocurrency market. For example, under a new law passed

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

161

investigation to determine if cryptocurrency exchanges and broker inter-

mediaries were engaging in the sort of conduct that regulators watch for

in financial markets, including whether exchanges were providing some

traders with superior access to data and speed.

263

The cryptocurrency ex-

changes that responded to the NYAG’s investigation self-reported they

were not.

264

Since the NYAG’s market investigation, securities regulators

globally have echoed the NYAG’s concerns.

265

In 2018, the International

Organization of Securities Commissions (IOSCO) released its own report

on the market, advising regulators to monitor whether cryptocurrency

by New York in 2015, exchanges operating in the state of New York have to sub-

mit their exchange rules to the state and obtain a license to operate. 23 CRR-NY,

ch. I, pt. 200. Under the Exchange Act, the SEC has jurisdiction over securities

and it has used the Howey Test to determine that virtual currencies are securities

and exercise jurisdiction over the market. Framework for “Investment Contract”

Analysis of Digital Assets, https://perma.cc/7WXS-95NG (last updated Apr. 3,

2019). The CFTC has ruled that Bitcoin and other virtual currencies are a “com-

modity” under Section 1a(9) of the Commodity Exchange Act of 1936 (CEA),

which grants the CFTC jurisdiction over market participants for fraud and ma-

nipulation. In re Coinflip Inc. et al., Respondent, CFTC, No. 15-29 (Sept. 17, 2015),

https://perma.cc/7F3G-ZY3L.

263

. UNDERWOOD, supra note 26.

264

. Id. at 25 (“Trading platforms that engage in proprietary trading on

their own venues uniformly claimed to the OAG that their trading desks had no

informational or other trading advantage over customers.”). Note, the NYAG

was nonetheless concerned that cryptocurrency exchanges provide varying lev-

els of access that may let sophisticated traders “leverage access and speed to

power sophisticated automated trading strategies – strategies that can negatively

affect the trading performance of everyday, non-automated customers” buying

and selling on the same venue. Id.

265

. BOARD OF THE INTERNATIONAL ORGANIZATION OF SECURITIES

COMMISSIONS, ISSUES, RISKS AND REGULATORY CONSIDERATIONS RELATING TO

CRYPTO-ASSET TRADING PLATFORMS: CONSULTATION REPORT (Int’l Org. of Sec.

Comm’n 2019) [hereinafter IOSCO CRYPTO CONSULTATION REPORT] (drawing on

feedback from securities regulators globally, including those in the U.S., the U.K.,

France, Germany, Japan and China, and raising concern over market integrity,

non-discriminatory access, management of conflicts of interest, and transpar-

ency); ASIA SECURITIES INDUSTRY AND FINANCIAL MARKETS ASSOCIATION, ASIFMA

BEST PRACTICES FOR DIGITAL ASSET EXCHANGES (2018) (raising similar concerns).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

162

actors are upholding principles long-held in the securities market, includ-

ing fair access to exchanges.

266

Given the similarities between the struc-

ture of the electronically traded equities and ads markets, ad exchanges

are deserving of the same attention from regulators and legislators.

B. Steps Toward Identifying and Managing Intermediary Conflicts of

Interest

1. Structural Separations

We might address other competition problems in advertising mar-

kets by leaning on a second core principle from the securities market

toolbox: identify and manage intermediary conflicts of interest. In finan-

cial markets, the approach to managing conflicts has ebbed and flowed

from requiring structural separations to imposing conflicts of interest and

disclosure rules. After the Great Depression, Congress passed the Glass-

Steagall Act, which took the structural separation approach and prohib-

ited banks, for example, from running a broker dealer or underwriting

securities offerings.

267

Financial conglomerates like J.P. Morgan had a

year to divest their conflicting operations (J.P. Morgan’s brokerage and

investment division became Morgan Stanley).

268

That structural separa-

tion approach later gave way to allowing some firms to engage in con-

flicting lines of business but requiring that they manage their conflicts.

269

266

. IOSCO CRYPTO CONSULTATION REPORT, supra note 265, at 10 (discuss-

ing need to “[e]nsure that access to the system or exchange and to associated

products is fair, transparent and objective” and that exchange procedures apply

fairly and on a non-discriminatory basis).

267

. Commercial banks were prohibited from engaging in investment

banking activity generally. See generally Banking Act, Pub. L. No. 73-66, 48 Stat.

162 (1933) (current version at 12 U.S.C. § 227).

268

. Julia Maues, Banking Act of 1933 (Glass-Steagall), FED. RES. HIST. (Nov.

22, 2013), https://perma.cc/2LUP-PATS (providing a summary of the structural

separation approach taken by Glass-Steagall).

269

. The Gramm-Leach-Bliley Act in 1999 repealed parts of the Glass-

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

163

Despite these relaxations, policymakers continue to prohibit a com-

pany that runs a public stock exchange from simultaneously running a

division that trades on the exchange. That division between the operator

of an electronic marketplace and the middlemen trading in the market is

upheld in other markets too. For example, the companies that operate the

major electronic event ticket marketplaces —Ticketmaster, Stubhub,

AXS, and Telecharge —do not simultaneously operate a ticket broker. To

be discussed further in the following Part, the centralized computerized

marketplaces for airfares also eventually parted ways with their airline-

owners. Google might similarly be required to divest its business divi-

sions that have a conflict with Google participating in the market as a

seller of ad space: its exchange, its buy-side and sell-side intermediaries,

and the Chrome browser.

2. Conduct and Disclosure Rules

Outside of such structural separations, we might manage conflicts of

interest through conduct and disclosure rules. In the equities trading

market, multi-service financial firms cannot misuse information belong-

ing to third-party traders, must maintain ethical walls, and cannot route

orders in a discriminatory manner.

270

Google’s ad trading intermediaries

(i.e., ad server & buying tools) might similarly be prohibited from abus-

ing their access to third parties’ sensitive information, be required to put

up ethical walls, and be prohibited from routing trading activity to

Google’s exchange and properties in a discriminatory manner.

Steagall Act to enable financial holding companies to underwrite securities, act

as a broker-dealer, and engage in other conflicting lines of business. Gramm-

Leach-Bliley Financial Modernization Act, Pub. L. No. 106-102, 113 Stat. 1338

(1999) (codified in scattered sections of 12 U.S.C. and 15 U.S.C.). See generally Jef-

frey Bingham, Scaling Chinese Walls: Insights from Aftra v. JPMorgan Chase, 4 WM.

& MARY BUS. L. REV. 767 (2013).

270

. For the best execution (routing) rule, see supra note 53. For conflicts

of interest data rules, restrictions on client data use, and ethical wall require-

ments, see supra notes 54-56.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

164

Lawmakers could also require Google’s ad trading intermediaries on

the buy-side and sell-side to act in the best interest of their customers. If

Google was required to do so, it would not have been able to negotiate

out of duties historically owed to its publisher and advertiser customers.

Such a fiduciary framework would revert ownership in ad server user

IDs back to publishers and advertisers, empowering them to share these

IDs as they see fit. Conversely, such a rule would prohibit Google from

merging data it obtains through the operation of trading intermediaries,

with other Google data, for the purpose of trading on inside information

advantages.

Importantly, countervailing user privacy concerns can be managed

by prohibiting companies from collecting sensitive consumer data (e.g.,

health), limiting what data can be used for targeted advertising, and giv-

ing consumers the ability to opt-out of behaviorally targeted ads alto-

gether through a simple, national, opt-out regime. This approach would

strike a healthier balance between privacy and competition.

Policymakers have ported this way of thinking about market struc-

ture to other electronic trading markets too. Take, for example, the grow-

ing concern over conflicts of interest and insider abuses in cryptocur-

rency markets. When the NYAG investigated the market in 2018, they

raised concerns that some cryptocurrency firms have conflicts: they are

operating an exchange while simultaneously trading in their own mar-

ket,

sometimes accounting for up to 20% of trading activity.

271

That find-

ing led the NYAG to flag the risk of “insider abuses” and advise that

“[m]anaging conflicts of interest is a serious and growing issue in the vir-

tual marketplace.”

272

271

. UNDERWOOD, supra note 26, at 25 (highlighting the risk of insider

abuses and noting that “[t]rading platforms [that responded to the Initiative] that

engage in proprietary trading on their own venues uniformly claimed to the

OAG that their trading desks had no informational or other trading advantage

over customers”).

272

. Id. at 21.

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

165

The international body of securities commissions echoed the same

concerns and advised that lawmakers take the regulatory “toolkit” for

managing conflicts and market abuses in financial markets and repur-

pose them for cryptocurrency trading.

273

In a recent report, it highlighted

that “[t]he existence of unmitigated conflicts can negatively impact inves-

tor protection and confidence, as well as fair, efficient and transparent

markets.”

274

The conflicts it highlighted included when a company runs

an exchange while simultaneously trading on it or possessing an interest

in a traded crypto asset.

275

Outside of the cryptocurrency marketplaces, lawmakers have exhib-

ited concern about conflicts of interest and associated “inside” infor-

mation abuses in gaming and betting markets. When it came to light in

2015 that an employee of major gaming platform DraftKings misappro-

priated information submitted by the platform’s users to inform their

own play to make a $350,000 profit, DraftKings was accused of allowing

“insider trading.”

276

The news spurred the NYAG to launch a probe into

273

. See generally IOSCO CRYPTO CONSULTATION REPORT, supra note 265

(highlighting concern over market integrity, transparency, and the management

of conflicts of interest).

274

. IOSCO MARKET STRUCTURE CONSULTATION REPORT, supra note 48, at

16 ( “CTPs that position themselves to provide end-to-end services . . . may have

additional conflicts. Traditionally, these roles have been performed by independ-

ent parties. When CTPs provide such end-to-end services, any conflicts of inter-

est that arise need to be mitigated to prevent potential market misconduct and/or

investor protection concerns.”).

275

. Id. at 20 (stating a key issue of IOSCO Principle 31 5(d) is “[a]ddress-

ing any conflicts of interest that arise between [a market intermediary’s] interests

and those of its clients. Where the potential for conflicts arise, a market interme-

diary should ensure fair treatment of all its clients by taking reasonable steps to

manage the conflicts through organizational measures to prevent damage to its

clients’ interest, such as: internal rules, including rules of confidentiality; proper

disclosure; or declining to act where conflict cannot be resolved”).

276

. Specifically, the employee used information about which players

DraftKings users picked for their roster. Joe Drape & Jacqueline Williams, Scandal

Erupts in Unregulated World of Fantasy Sports, N.Y. TIMES (Oct. 5, 2015),

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

166

whether employees could “gain[] an unfair financial advantage in a con-

test … by exploiting their access to nonpublic data.”

277

DraftKings rival

FanDuel then started prohibiting its own employees from playing fan-

tasy sports games at all, as well as prohibiting employees of other plat-

forms from playing on its site.

278

Similarly, after a Supreme Court ruling

in 2018 permitted individual U.S. states to legalize sports betting, law-

makers introduced federal and state legislation prohibiting bets made on

material nonpublic information (colloquially called “insider betting”).

279

The State of New Jersey passed legislation prohibiting operators of sports

betting operations from taking wagers from persons that work for or are

affiliated with the operator.

280

https://perma.cc/LCG7-GCL6 (where sports and gambling attorney Daniel Wal-

lach remarks, “It is absolutely akin to insider trading. . .It gives that person a

distinct edge in a contest.”); REUTERS, DraftKings and FanDuel on Defensive over

Insider Betting Revelations, GUARDIAN (Oct. 5, 2015), https://perma.cc/G3XH-

RMU8; Bourree Lam, The New York Attorney General's Investigation of DraftKings

and FanDuel, ATLANTIC (Oct. 7, 2015), https://perma.cc/V8T9-LVNA/.

277

. Lam, supra note 276; Darren Rovell, New York Attorney General

Launches Inquiry into DraftKings, FanDuel, ESPN (Oct. 6, 2015),

https://perma.cc/GG4P-99YQ. The New York Attorney General’s office later

brought and settled suit with DraftKings and FanDuel for false and deceptive

advertising. Press Release, Eric Schneiderman, Att’y Gen., N.Y., A.G. Schneider-

man Announces $12 Million Settlement with DraftKings and Fanduel (Oct. 25,

2016), https://perma.cc/C6QM-LY4Z.

278

. Drape and Williams, supra note 276.

279

. Murphy v. National Collegiate Athletic Association, 138 S. Ct. 1461

(2018). For an example of legislation introduced at the state level, see, e.g., S17D

§ 1367(w)(vii) (N.Y. 2019) (defining a “prohibited sports bettor [as] . . . any indi-

vidual with access to non-public confidential information about sports wager-

ing” among other circumstances”). For legislation introduced at the federal level,

see Sports Wagering Market Integrity Act of 2018, S. 3793, 115th Cong. (2018).

280

. Assemb. B. 4111 § 2(f)(4) (N.J. 2018) ( “[A]n operator shall adopt pro-

cedures to prevent persons from wagering on sports events who are prohibited

from placing sports wagers. An operator shall not accept wagers from any person

whose identity is known to the operator and . . . who has access to nonpublic

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

167

Policymakers similarly managed intermediary conflicts of interest in

the computerized marketplaces for airline tickets.

281

In the late 1970s, air-

lines began routing their airfares into centralized marketplaces, now

called global distribution systems (GDSs), where travel agents could find

and book airfares electronically.

282

In the early stages of this market’s evo-

lution, major airlines like American and United owned and ran the main

GDSs, Sabre and Apollo, respectively—similar to how Google, as the

largest seller of digital advertising today, concurrently runs the largest

advertising exchange.

283

Early on, rival airlines lodged complaints over

how the airline-owned marketplaces were preferentially routing travel

agents to the airline-owner’s tickets.

284

As a result, the Civil Aeronautics

Board (CAB) and the DOJ opened investigations in 1982 and 1983.

285

confidential information held by the operator.”). Outside of introduced legisla-

tion, the media has continued to explore this parallel between information ad-

vantages in sports betting and insider trading. See Jacob Gershman, The Brave

New World of Betting on Athletes’ Data, WALL ST. J. (March 10, 2020),

https://perma.cc/WG7U-D84H (discussing how athletes’ biometric health data

could soon provide unfair information advantages when sports betting).

281

. For a history of the early evolution of the airline intermediary market,

see Duncan Copeland & James McKenney, Airline Reservations Systems: Lessons

from History, 12 MGMT. INFO. SYS. Q. 353 (1988).

282

. The GDSs (initially called CRSs) were intermediary markets between

airlines and travel agents. See Cindy Alexander & Yoon-Ho Lee, The Economics of

Regulatory Reform: Termination of Airline Computer Reservation System Rules, 21

YALE J. ON REG. 369 (2004).

283

. See Douglas Feaver, CAB Chairman Promises Best Computer Ticket In-

formation, WASH. POST (July 21, 1983), https://perma.cc/RAR3-LVTN (“American

and United airlines both sell computer reservation systems and between them

have locked up about 80 percent of the market.”); Aimee Minick, Comment, Com-

puter Reservations Systems, Airlines, and the Internet, 65 J. AIR L. & COM. 891 (2000);

Computer Reservation Systems: Alleged Competitive Abuses and Consumer In-

jury, 48 Fed. Reg. 41,171, 41,173 (Sept. 14, 1983) (codified throughout 14 C.F.R.

ch. II).

284

. See generally Copeland & McKenney, supra note 281.

285

. Id. at 363 (summarizing early investigations by CAB and the DOJ).

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

168

After lengthy investigations, the CAB and the DOJ concluded these

practices distorted competition.

286

This kickstarted two decades of regu-

lation of these computerized marketplaces starting in 1984, even though

the wider airline industry was deregulated in 1978.

287

To manage conflicts

of interest, the regulatory regime sought to correct problems of “fair ac-

cess.”

288

The first rules promulgated by CAB in 1984 prohibited airline-

owned-marketplaces from steering travel agents to the GDS-owner’s

own tickets and engaging in other forms of discrimination.

289

286

. U.S. DEP’T. OF JUST., REPLY COMMENTS TO NOTICE OF PROPOSED

RULEMAKING COMPUTER RESERVATION SYSTEM REGULATIONS (2003) (explaining

how this practice was exclusionary); Larry Locke, Flying the Unfriendly Skies: The

Legal Fallout Over the Use of Computerized Reservation Systems as a Competitive

Weapon in the Airline Industry, 2 HARV. J. L. & TECH. 219, 224 (1989) (summarizing

that the DOJ ultimately did not bring charges but did find that airlines used their

control of CRSs to weaken competition in the airline market); Alexander & Lee,

supra note 282, at 380 (discussing how CAB thought preferential routing fore-

closed competition and how these concerns underpinned CAB rules promul-

gated in 1984).

287

. The Airline Deregulation Act of 1978 ended forty years of the Civil

Aeronautics Board (CAB) regulating the airlines as a public utility (e.g., CAB set

routes and fares). The Airline Deregulation Act also vested ongoing regulatory

authority in CAB until 1984 and Section 411 of the Federal Aviation Act permit-

ted the DOT to subsequently assume regulatory oversight of the GDS market.

Note, however, Congress gave the DOT a mandate to prohibit unfair methods of

competition, which captures a wider range of conduct than that prohibited under

antitrust laws. See generally Timothy Ravich, Deregulation of the Airline Computer

Reservation Systems (CRS) Industry, 69 J. AIR L. & COM. 387 (2004); Alexander &

Lee, supra note 282.

288

. See, e.g., Feaver, supra note 283 (where Dan McKinnon, Chairman of

the Civil Aeronautics Bureau, states that the forthcoming rule “would take a

close look at ensuring fair access” for airlines and consumers).

289

. As a part of these rules, the GDSs could not price discriminate be-

tween airlines, had to charge airlines the same fees for the same levels of service,

had to make certain GDS data available to all participating airlines for purchase,

and could not engage in display and functionality bias. Carrier-Owned Com-

puter Reservations Systems, 49 Fed. Reg. 11,644 (Mar. 27, 1984) (codified at 14

C.F.R. pt. 255); Alexander & Lee, supra note 282, at 395-96 (providing a summary

of the non-discrimination rules).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

169

The Department of Transportation (DOT) later took over the regula-

tion of these marketplaces and dealt with a second discriminatory rout-

ing problem similar to one that Google presents today. American and

United airlines refused to make their tickets available for sale in rival

marketplaces, just as Google today refuses to sell its Search and YouTube

ad space through rival exchanges and buying tools.

290

Because an airline

like American had power in particular geographic airfare markets, it

could pressure travel agents in those locations to use its own market-

place, Sabre, instead of a competitor’s.

291

But once agents chose Sabre, Sa-

bre would steer travel agents to buy American airfares over others.

292

The

initial “withholding” permitted a major airline like American to distort

competition in the intermediary airfare marketplaces, as well as in the

airfare market, because travel agents tended to use just one reservation

system.

293

The architecture of the advertising market presents the same

concerns.

To address competition concerns, the DOT adopted a “mandatory

participation rule” in 1992 that forced airlines like American and United

to route their inventory into rival ticket marketplaces in addition to their

own.

294

Wanting to stop the dominant European airlines from similarly

290

. See generally Alexander & Lee, supra note 282, at 380 (explaining the

different ways airlines distorted competition between reservations systems, in-

cluding by refusing to “participate in” rival computer reservation systems).

291

. The withholding of tickets from rival GDSs was also thought to con-

stitute a barrier to entry in the GDS market. See generally id. at 393.

292

. Id. at 379-80.

293

. Computer Reservations System (CRS) Regulations: Statements of

General Policy, 67 Fed. Reg. 69,366 (Nov. 15, 2002) (codified at 14 C.F.R. pts. 255,

399), (stating that travel agents primarily used one reservation system, which re-

sulted in the chosen system having market power over agents).

294

. The 1992 mandatory participation rule required airlines that operated

a marketplace to route their tickets into rival marketplaces as long as terms were

commercially reasonable. Computer Reservation System (CRS) Regulations, 57

Fed. Reg. 43,780, 43,800 (Sept. 22, 1992) (“We conclude that this [mandatory par-

ticipation] requirement is justified on competitive grounds, since it will keep a

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

170

distorting competition, Europe adopted mandatory routing out rules

too.

295

However, attempts to manage conflicts of interest eventually gave

way to structural separations. Facing continued regulatory scrutiny, the

airlines eventually divested their ownership interests in the GDSs alto-

gether.

296

Google might similarly be required to route YouTube ad space out to

rival exchanges and buying tools. However, overall, the complexity and

dynamism of high-tech ad markets, combined with the difficulty law-

makers had enforcing non-discrimination rules in the airfare market-

places and the challenges financial regulators continue to have policing

ethical wall separations, likely buttress efficacy arguments for the struc-

tural separation approach.

297

CRS owner from using its dominance of a regional airline market as a tool for

obtaining dominance in the area's CRS market.”); see generally Alexander & Lee,

supra note 282. A parallel issue comes up in other markets. For example, the New

York Attorney General’s office recently stated that if NFL sports teams require

ticketholders to resell their NFL tickets using a specific ticketing market or ex-

change, this conduct might be a “restraint of trade” that “impede[s] consumer

access to alternative ticketing resale platforms.” Schneiderman, supra note 26, at

32.

295

. Council Regulation 2299/89 of July 24, 1989, On a Code of Conduct

for Computerized Reservation Systems, O.J. (L 220) 1 (EEC) (prohibiting CRSs

from engaging in discriminatory display or routing and requiring CRSs to permit

all airlines to participate in a non-discriminatory manner).

296

. Ravich, supra note 287 (discussing airline divestiture of GDS owner-

ship interests and concomitant deregulation); Dawit Habtemariam, A Brief His-

tory of Air Travel Distribution, BUS. TRAVEL NEWS (Oct. 28, 2018),

https://perma.cc/C7WL-53V2 (“At the same time, airlines were divesting their

ownership in CRSs. Sabre launched an IPO in 1996 and became fully independ-

ent of American Airlines in 2000. Galileo, which had merged with Apollo, went

public in 1997. Amadeus, which had absorbed System One, went public in 1999.

. . . Given the growing competition and the fact that airlines were divesting their

ownership in CRSs, the DOT eliminated some of its CRS regulations and let the

rest expire.”).

297

. See, e.g., Khan, supra note 20 (arguing that structural separations are

preferable to non-discrimination rules because they may be more effective in po-

licing conduct and they reduce regulatory burden).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

171

3. Transparency and Disclosure

More transparency and disclosure would also advance healthier

competition in advertising markets. In the securities market, mandatory

disclosure requirements allow market participants to more fully under-

stand how the complex trading markets work and to better monitor in-

termediary conflicts of interest.

298

Financial exchanges must disclose to

regulators how their trading operations work and obtain regulatory ap-

proval for their exchange operations. Broker dealers and other financial

intermediaries must disclose information about their trading activity (in

milliseconds or a finer increment) and even synchronize business clocks

with a universal clock to let others monitor whether they are properly

managing their conflicts of interest (and not front-running).

299

Transpar-

ency along the same lines in advertising markets would effectuate the

same ends.

The principles of transparency and disclosure have been utilized in

the emerging cryptocurrency markets too. For example, the State of New

York recently passed legislation requiring cryptocurrency exchanges to

disclose information about their trading operations before obtaining a li-

cense to operate as an “exchange.”

300

When the NYAG investigated this

298

. See, e.g., INT’L ORG. OF SEC. COMM’N, OBJECTIVES AND PRINCIPLES OF

SECURITIES REGULATION (2003) (discussing how transparency is a core objective

in securities regulation and how transparency helps police market manipulation

and unfair trading practices); see also Julie Manasfi, Systemic Risk and Dodd-Frank’s

Volcker Rule, 4 WM. & MARY BUS. L. REV. 181 (2013) (discussing how disclosure

rules can help to manage conflicts of interest).

299

. Reporting requirements in Exchange Act Section 13(d) and SEC Rule

13d-1 require brokers to disclose information about proprietary trades in Sched-

ule 13D. 17 C.F.R. § 240.13d-1; 17 C.F.R. § 242.613 (requiring all national securities

exchanges and associations to file a “national market system plan to govern the

creation, implementation, and maintenance of a consolidated audit trail and cen-

tral repository” to collect and keep relevant data on trading practices); BD. OF THE

INT’L ORG. OF SEC. COMM’N, CLOCK SYNCHRONISATION: CONSULTATION REPORT

(2019).

300

. See discussion supra note 262.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

172

market in 2018, it specifically investigated the issue of transparency

around fees, noting “[f]ee transparency is especially important in a com-

plex electronic trading environment like virtual currency.”

301

However,

one thing that distinguishes cryptocurrency markets from advertising

markets is how cryptocurrency exchanges have already created self-reg-

ulatory organizations (SROs) to help to facilitate the surveillance of ma-

nipulative and fraudulent trading activity.

302

Advertising markets would benefit from the same transparency and

disclosure approach. Google has pushed back on disclosure requirements

for “privacy” reasons, but it is important to note that financial regulators

successfully balance similar privacy concerns.

303

At the end of the day,

lawmakers strike the right balance with disclosure requirements that al-

low market participants to understand the market and police intermedi-

aries, alongside industry specific rules that simultaneously address pri-

vacy.

304

V.CONCLUSION

Approximately 86% of online display ad space is now bought and

sold on electronic trading venues, where buyers and sellers must go

through a computerized intermediary to trade. Google dominates these

301

. UNDERWOOD, supra note 26, at 11.

302

. VIRTUAL COMMODITY ASSOCIATION, https://perma.cc/7MVF-KGBC

(last visited Nov. 18, 2020).

303

. Press Release, Jay Clayton, Chairman, SEC, Statement on Status of the

Consolidated Audit Trail (Sept. 9, 2019), https://perma.cc/45ZA-CHVF; Patrick

Temple-West & Robert Armstrong, ‘Hubble Telescope for Markets’ Attracts Criti-

cism, FIN. TIMES (Oct. 29, 2019), https://perma.cc/DF64-ZQ9J (Dennis Kelleger, of

consumer advocacy group Better Markets, stating “The finance industry does not

want to [sic] the regulators to have the ability to monitor and police their behav-

iour, and they are using privacy as a pretext.”).

304

. This delicate balance came up recently with the SEC’s consolidated

audit trail, which is a comprehensive market surveillance initiative. See Clayton,

supra note 303 (discussing the balancing of personally identifiable consumer in-

formation and market surveillance).

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

173

markets and distorts competition by engaging in conduct that lawmakers

have experience with in other electronic trading markets. Google reserves

for itself speed and information advantages, routes order flow to its ex-

change and websites in a discriminatory manner, and breaches ethical

walls to use the sensitive, nonpublic information belonging to third-party

buyers and sellers to inform its own trading activity. To protect competi-

tion in advertising, lawmakers might borrow the core principles financial

regulators have already crafted to address these types of competition

problems in other electronic trading markets. That is, require exchanges

to provide all traders with non-discriminatory access to information and

speed, identify and manage intermediary conflicts of interest, and require

trading disclosures to advance these principles in order to protect the

overall integrity of the market.

STANFORD TECHNOLOGY LAW REVIEW Vol. 24:1

174

VI.APPENDIX

A. Timeline

Fall 2020 WHY GOOGLE DOMINATES ADVERTISING MARKETS

175

B. Screenshot of Auction Timestamp Transparency

Image of header bidding timeouts on The Washington Post on No-

vember 30, 2019.