Case Study: Google

Google is the world’s largest search engine with approximately 92.19 percent of the global search engine market.¹ The platform currently ranks first for global internet engagement on Alexa rankings,² meaning that it is the single most visited website in the world.³ Google is also the largest player in the digital advertising market, with a 37.2 percent market share.⁴ Advertising funds the majority of Google’s, and its parent company Alphabet’s, operations, with about 85 percent of Alphabet’s annual revenue coming from ads.⁵

Although Google was not the first digital advertising platform, it is responsible for establishing targeted advertising as a key industry practice.⁶ In doing so, the platform reaped massive financial rewards and laid the groundwork for surveillance capitalism to become an established practice throughout the technology industry.⁷ In the late 1990’s, Google engineers recognized the vast potential of the continuous flows of behavioral data that could be inferred from the company’s expansive data logs from search queries. Google initially used this data to refine and improve its product offerings. In late 2000, Google also harnessed this data to begin matching digital ads with queries, with the aim of delivering relevant ads to users and ensuring value for advertisers.⁸ The insights provided by this behavioral data were paramount, and Google eventually also used this data to predict user behavior, which further spawned the growth of the company’s ad business.⁹

Like many other online platforms and digital advertising platforms, Google collects and monetizes a staggering amount of user data, including personal information on users such as their age and location, which is used to target and deliver ads to them across the internet. It collects this information based on factors such as what a user searches for using its Search product, the videos they watch on YouTube, and the places they look for on Google Maps.¹⁰ The company also collects sensitive information about individuals even when they are logged out of their Google accounts, browsing in a private browsing mode, or not even Google account owners in the first place. All of the information that Google collects on individuals online is used to target and deliver ads to them across the internet.

In July 2018, Google rebranded and reshuffled its ad products so that they focused on three core offerings: Google Ads, Google Marketing Platform, and Google Ad Manager. Google Ads is a rebranded version of AdWords, a search-based paid advertising system. The Google Marketing Platform merges Google Analytics 360 Suite and DoubleClick Digital Marketing. Google Ad Manager is a programmatic advertising platform that merges DoubleClick for Publishers and DoubleClick Ad Exchange. These will be discussed in more depth in the next section.¹¹

The primary mechanism that Google uses for targeting ads is DoubleClick, an advertising delivery and tracking company that Google acquired in 2007. As previously mentioned, portions of DoubleClick were merged into the new Google Marketing Platform and Google Ad Manager in June 2018. Therefore, mentions of DoubleClick refer to its technology, as the actual brand has been distributed throughout Google’s platform.

DoubleClick relies on web cookies to track and collect information on users’ browsing behaviors. It primarily does this by monitoring user IP addresses. This information can be used to infer a great deal about users’ behaviors, habits, and interests, which is then used by DoubleClick to target and deliver ads to users.¹² However, DoubleClick does not have access to a user’s real identity. Information such as users’ names, email addresses, and search histories lies with Google proper. ¹³ When a user visits a web page that hosts advertising content through Google’s Display Network or its DoubleClick Ad Exchange (now part of Google Ads and Google Ad Manager respectively), DoubleClick is able to identify the users’ digital identity and determine which ads the user should see.¹⁴ This is possible even if the user does not arrive at this web page through a Google product or service such as Google Chrome or Google Search.

The primary mechanism through which businesses can deliver and place ads on Google is through the Google Ads platform. Google AdWords was launched in 2000 and rebranded as Google Ads in 2018. Google Ads enables advertisers to create and deliver mobile and desktop ads. These ads can appear on Google’s search engine results page (SERP), on other Google products and platforms (e.g. YouTube and Blogger), and on websites that opt in to having Google-placed ads on their web pages.

Using Google Ads, an advertiser can run three types of campaigns—search ad campaigns, display ad campaigns, and video ad campaigns. Search ads are text-based ads that appear alongside search results on the SERP. Display ads are conventionally image-based ads and they appear on websites that opt in to having Google place text- and image-based ads on their web pages. These websites opt in to this process by becoming part of an ecosystem of such websites, known as the Google Display Network (GDN). A website that is interested in hosting online ads must install software onto their website that relays information about a current user, such as their search criteria, to Google. Google will then place ads it deems relevant on their website. Through this process, these host websites receive a cut of the money from the ads. When an ad placed by Google appears on a website other than Google.com, it will be labeled with the banner “Ads by Google.”¹⁵ The third category of ads, video ads, typically last between six and 15 seconds and appear on YouTube before, during, or after a video plays.¹⁶

There is no upfront cost for advertisers who want to run ads using the Google Ads platform.¹⁷ The profit model of Google Ads is based on three types of bids. The first is cost-per-click (CPC), in which an advertiser pays for every click their ad receives. The second is cost-per-mile (CPM) in which an advertiser pays for 1000 ad impressions. The third type of bid is cost-per-engagement (CPE) in which an advertiser pays a certain amount when a user demonstrates a predicted action with an advertiser’s ad. When a user clicks or engages with an ad on a website that is part of the GDN, Google and the website owner split the profits.¹⁸ Websites that are part of the GDN can be paid per-click or per-impression for the ads that appear on their web pages.¹⁹

When an advertiser wants to place a bid for an ad campaign, they can do so using an automated or manual bidding process. Manual CPC bidding lets advertisers set their own maximum cost per click for their ads. Advertisers can set a default bid amount for their entire ad group. They can also set separate bid amounts for individual ad keywords or placements, which they may have identified as more profitable. As an ad campaign runs, Google Ads will help advertisers evaluate and optimize their performance.

Advertisers who want even more detailed information can use Google Analytics.²⁰ Google Analytics is a Google product that can track and report on website traffic to an ad.²¹ It also feeds data back to Google, which can use the data to enhance its targeting practices and parameters.²² According to the 2016 Englehardt and Narayanan study, Google Analytics is present on almost 70 percent of websites.²³ During automated bidding, an advertiser can choose from a number of automated bidding strategies, such as maximize clicks, maximize conversions, and maximize conversion value. Conversion is when a user engages with an ad and then performs an action an advertiser deems valuable such as making a purchase or communicating with the business.²⁴

During the automated bidding process, an advertiser will set a maximum budget that they are willing to spend. Google will take into account this maximum budget and adjust an advertiser’s bid based on competing bids in order to optimize the advertiser’s selected bid strategy.

A bid is not the only factor that influences who wins an ad auction, however. Google determines how to rank and place ads by calculating an Ad Rank value. The higher an advertiser’s Ad Rank value, the higher its ads will rank on the list of ads to place.²⁵ Ad Rank is calculated by multiplying the maximum bid placed by an advertiser with an advertiser’s Quality Score (QS). The QS is an estimate of how relevant an advertiser’s ads, keywords, and landing page are to a user who sees an advertiser’s ad. A higher QS typically results in lower costs for an advertiser and a better ad position.²⁶ This is because a user who engages with relevant content is more likely to continue using Google’s services. As a result, Google is not only interested in the price advertisers pay, but it is also in Google’s business interests to ensure relevant ads are displayed to users.²⁷ The QS is calculated by measuring the quality of an ad using a range of factors. These include:

1. The Landing Page Experience: An estimate of how relevant and useful an advertiser’s landing page is to users who click on their ad. This metric accounts for how well the content on an advertiser’s landing page matches a user’s search query and how easy it is for users to navigate the landing page.
2. Expected Clickthrough Rate (CTR): A measure of how likely it is that an advertiser’s ad will be clicked on when shown. This score is based on the past clickthrough performance of an advertiser’s ads. It does not consider factors that could affect the visibility of an ad, such as ad position, and extensions.
3. Ad Relevance Measure: A measure of how closely an advertiser’s keyword matches the message in their ads. A below average ad relevance score can indicate that advertiser’s ads are too general or too specific to answer a user’s query, or that the keyword is not relevant to the advertiser's business.²⁸ The higher an ad’s QS, the higher it will be ranked and placed online.²⁹

Advertisers can also opt to use responsive search ads. This is when an advertiser provides Google with a set of 15 headline texts and four variations of ad copy. Over time, using its analytics and insights, Google Ads will automatically test different combinations of these headlines and ad copy³⁰ using A/B testing. It will then identify, select, and run which of these combinations performs the best.³¹

Google AdSense is the primary mechanism that online publishers such as bloggers or website owners can use to host ads on Google. The AdSense platform is distinct from the Google Ads platform and was launched in 2003. It enables contextual advertising by letting online publishers opt in to allowing Google to place ads on their websites, which are targeted to the publishers’ content or audience.³² When a user enters search criteria into a website that is using AdSense, AdSense will place ads that it deems relevant to the user’s search on the search results web page. If an advertiser, also known as a “sponsor,” is interested in placing an ad online using AdSense, they provide Google with search criteria and potential ads that Google can deliver if a relevant match occurs. Sponsors will also provide Google with a bid they are willing to pay if a reader ends up clicking on an ad that was delivered to them.³³

In order to determine which ads should be displayed to a user during a search query, Google runs an automatic bid auction in real-time for the same search criteria. Typically, the ad with the highest bid will be displayed first, and the remaining ads will be ranked sequentially. If a bid is too low or if displaying an ad will result in passing a threshold related to an advertiser (e.g. exceeding the maximum account total for the sponsor), then Google may not show the ad.³⁴ AdSense automatically places targeted ads, which are created and paid for by advertisers, on a participating publisher’s website.³⁵ Google pays these web publishers based on the number of clicks or impressions ads receive.³⁶ Since advertisers pay different prices for different ads, the amount that a web publisher can earn through using AdSense varies.³⁷

According to a 2016 study by Englehardt and Narayanan, DoubleClick is present on almost 50 percent of websites³⁸ thanks to “cookie syncing,” which enables different trackers to share user identifiers with each other. This makes DoubleClick one of the most promiscuous cookie-syncing third parties, sharing 108 different cookies with 118 other third parties.³⁹

Google’s ad targeting and delivery practices have been scrutinized for their ability to produce harmful results. Many researchers have documented how the platform’s targeting parameters can enable advertisers to generate discriminatory outcomes. As previously noted, the process of establishing targeting parameters involves discriminating or differentiating among categories of users as it requires advertisers to include and exclude users based on their personal traits, behaviors, and preferences. While not all types of discrimination against users is problematic, in cases where the discrimination is based on protected classes or other sensitive categories, this can result in harmful and even illegal outcomes that negatively affect protected classes and exacerbate hidden societal biases.

According to a Google spokesperson, Google’s ad targeting policies prohibit targeting based on traits such as race, ethnicity, religion, disability, and negative financial standing. Despite this formal policy, however, advertisers are able to target ads based on related parameters such as address, which can be a proxy for race due to segregated housing patterns.⁴⁰ In addition, as described below, even beyond deliberate use of proxies, targeted ads may lead to discriminatory patterns by detecting and perpetuating hidden biases in underlying data.

For example, in 2017 BuzzFeed News revealed that Google’s ad platform permitted advertisers to target users using discriminatory terms. In addition, the investigation found that Google’s ad platform also suggested discriminatory keywords in response to such harmful terms. These suggestions were generated by automated tools. For example, entering “white people ruin” as a potential advertising keyword returned automated suggestions to run ads next to searches including “black people ruin everything.” Similarly, entering “why do Jews ruin everything” returned automated suggestions to run ads next to searches including “the evil jew” and “jewish control of banks.”

Following an inquiry to the company about the keywords Buzzfeed used in its investigation, Google disabled all of the keywords except “blacks destroy everything.” The company stated that just because a keyword is eligible, it does not mean an ad campaign will necessarily be run using it. In addition, the company shared that it has language that informs advertisers when their ads are offensive and subsequently rejected. However, given they failed to catch these particular keywords, this investigation revealed how the use of automated tools during the ad targeting phase can result in the exacerbation of societal biases and discriminatory outcomes.⁴¹

Numerous researchers have also identified instances of discriminatory outcomes during the ad delivery phase while using Google’s ad platform. A 2013 study on discrimination in online ad delivery found that when the researcher conducted a Google search for a person’s name, Google AdSense would return personalized ads for public records on the SERP. The study found that when the search terms included names that were commonly assigned to African American children, the resulting public record ads suggested that the individual had an arrest record 81 percent to 86 percent of the time. When the search terms included names that were primarily assigned to white children, the word “arrest” typically did not appear in the text of the public records ads that were delivered.

This study provided a dramatic example of how the algorithmic decision-making of Google AdSense’s delivery system detected and perpetuated societal biases. Although historical bias in the American criminal justice system has resulted in disproportionate arrest rates for African Americans, the AdSense algorithm was also detecting and perpetuating the societal notion that African American names would be associated with an arrest record in public records.⁴² In fact, however, in many of these cases, although the ads were suggestive of an arrest record, after actually clicking through, a user would find that these individuals had no actual arrest record.

In addition, a 2015 study found that when users self-identified their gender in their ads settings, Google was less likely to show female job seekers ads for coaching services for high paying jobs, compared to male users. In order to test this, the researchers built an automated system that mimicked 17,370 jobseekers, and indicated the gender of each fake user in the ad settings.

The profiles were then shown 600,000 ads that the researchers tracked and evaluated. The study highlighted how Google’s ad delivery system made inferences about what types of ads were relevant to a user based on the user’s identified gender. This raised significant concerns, as these results reflected existing societal biases and highlighted how such tools could be used to further exacerbate such biases, by for example, furthering the gender pay gap.⁴³

In some cases, AdSense has also raised issues when advertisers have found their content placed next to controversial or harmful content such as hate speech. This often results in public backlash as well as advertisers pulling their ads from Google’s ad platform.⁴⁴ It has also highlighted the need for brand safety, and advertisers are increasingly looking to advertising platforms to ensure that their ads are placed next to brand-safe content.

Given the range of controversies Google has faced related to its ad practices, the company has come under significant pressure to provide greater transparency and accountability around its ad operations. The company has taken some steps to remedy these issues. However, given the vast reach and influence of its tools, and because often times discriminatory outcomes can result from societal biases that are hidden in the data that Google’s algorithms are processing, efforts to date have been insufficient to address these serious and complex problems.

Google’s ad targeting and delivery practices have also raised numerous privacy concerns. In the last several years, Google has varied its practices regarding what data it collects and combines for purposes of ad targeting. Prior to 2016, Google explicitly stated that unless users opted-in, the company would not combine users’ personal information collected from Google products such as Gmail and YouTube with the information it collected on users’ browsing activity, such as IP address and search queries,⁴⁵ using DoubleClick.⁴⁶ Despite this, the company introduced this practice in June 2016. Google could now deliver personalized ads to users based on data such as the keywords⁴⁷ they used in Gmail or receipts they had in their Gmail account.⁴⁸ This change raised a number of privacy concerns and signaled a significant increase in the company’s invasive data collection practices.⁴⁹ In June, 2017, the company announced that it would end this practice before the end of that year, and Gmail content would not be used or scanned for any ads personalization.⁵⁰

Google also buys data on consumers’ credit card purchases and other offline activity in order to enhance its ad targeting and delivery offerings.⁵¹ The company asserts that by harnessing all of these data points, it will be able to deliver users with ads that are more useful and relevant to their interests.

Following the controversy around the use of political advertising to influence and suppress voters during the 2016 U.S. presidential election, Google created a database of paid political advertising from its ad platform in its transparency report. The database includes information on political ads in the United States, the European Union, and India. Information in the database includes the total ads run and the total ad spend since May 31, 2018, ad spend by geography (by state or country), the top advertisers, and the top keywords. This data is updated weekly and is available for download in a CSV (comma separated values) file. In addition, each ad in the database is accompanied by information on the ad’s format (e.g. video), the amount spent on the ad, a range of the impressions the ad accrued, and a date range that the ad ran in.

Although these data points are a valuable starting point, the database does not include granular information on how and to whom these ads were targeted and the reach and engagement they had. For example, Google currently provides a range figure of the impressions an ad received, but this does not include detailed engagement information such as the number of likes, shares, and video views the ad received. In addition, the database does not include information on the size and characteristics of the user categories that the ad was delivered to. Transparency around this information is important because it is often required for researchers in order to identify instances of discrimination.

Further, Google’s database and transparency report only cover three regions—the United States, the European Union, and India. In December 2019, the company expanded its election advertising transparency in the United States to include state-level candidates and officeholders, ballot measures, and ads that mention federal or state political parties.⁵² However, these transparency efforts are limited to certain markets, and are not applicable to all of Google’s users and ad operations around the world, despite the fact that these services are used extensively globally.⁵³

As a result, this database falls short in its attempts to provide meaningful transparency and accountability around Google’s political advertising operations to all of its users.

Additionally, Google does not provide a comprehensive database of other types of ads outside political ads on its platforms, even though they may also yield discriminatory outcomes. In March 2019, the Department of Housing and Urban Development (HUD) charged Facebook for discriminatory ad practices related to housing. Shortly thereafter, HUD announced it would also be investigating Google and Twitter’s ad practices, suggesting that discriminatory advertising around housing may also be a concern on Google’s ad platform. Investing resources into expanding this ad transparency database to include all categories of ads would therefore be in the best interest of both the public and the company.

In addition, Google’s transparency report does not provide any data on how the platform enforces its ad content and targeting policies, and how many ads and accounts are subsequently removed or suspended. This demonstrates a significant gap in transparency and accountability around how the platform’s ad content and targeting policies are enforced. In December 2019, France’s competition watchdog, l’Autorité de la Concurrence, fined the company €150 million for using and randomly enforcing “opaque and difficult to understand” operating rules on its ad platform. The authority has ordered the company to provide further clarity around how it creates its Google Ads rules and how it enforces these rules to suspend accounts.⁵⁴

Google has taken some limited steps to address concerns regarding transparency and accountability around their ad targeting practices. In November 2019, the company announced that it would limit targeting options for verified advertisers seeking to run election ads so that these advertisers can only target users based on geographic location (down to the postal code level), age, and gender. Advertisers will be able to continue targeting users with contextual targeting (e.g. delivering ads to users reading or watching a story about the economy). The policy change applies to ads on Google Search, YouTube, and through the GDN. Google initially rolled this policy out in the United Kingdom ahead of the General Election in December 2019, and then expanded it to the European Union at the end of 2019, and globally in the beginning of 2020.⁵⁵

As the debate around appropriate rules for political ads online continues, many civil society organizations and researchers are calling for internet platforms to ban political ads altogether. In March 2019, Google banned political ads in Canada ahead of the country’s federal elections. This followed Canadian lawmakers’ enactment of a bill in December 2018 that will go into effect in June 2020 and will require online platforms to keep a record of all political and partisan ads they publish directly or indirectly during an election season. According to Google, complying with this law would necessitate major alterations to its advertising system, as it deploys a real-time bidding procedure. Given these technical barriers and the fact that both political and partisan ads are hard to define and thus hard to monitor, Google decided to ban political ads in the country rather than risk non-compliance, which could result in fines or jail time.⁵⁶

Other prominent voices, such as Ellen L. Weintraub, the chair of the Federal Election Commission, have suggested that instead of banning political ads altogether, platforms should prohibit the practice of microtargeting political ads. Microtargeting is a marketing and advertising strategy that harnesses user data, such as their interests, online connections, purchase history and demographic data, in order to segment users into smaller, more precise groups for ad and content targeting.⁵⁷ According to Weintraub, the practice of political ad microtargeting, enables advertisers to identify vulnerable groups and deliver political misinformation to them with little accountability, as these ads are never seen by the larger public.⁵⁸ Numerous researchers, technologists, and civil society and advocacy organizations have joined this call for a prohibition on political ad microtargeting.⁵⁹ Others, however, have pushed back, stating that limiting microtargeting prevents grassroots candidates from establishing themselves and puts them at a disadvantage.⁶⁰

Further, following controversy over Facebook’s decision to exempt political ads from its fact-checking policy in October 2019, Google clarified that it prohibits making false claims in political ads.⁶¹ However, some political leaders in the United States have criticized the company for limiting this ban only to false claims that could result in voter suppression or threaten election integrity, and not applying it to cover false claims regarding specific candidates.⁶²

Google has extended some limited features to its users in order to give them greater control over their ad experience. For example, Google users can adjust certain ad preferences and settings through the Google Accounts page. This includes being able to turn ad personalization off. It also includes being able to stop the platform from using activity and information from Google services to personalize ads. This applies to ads on websites and apps that partner with Google to show ads. When enabled, this feature stores data from these partner websites and apps in a user’s Google Account. This page also includes information on how a user’s ads were customized, and it explains that ads are customized based on personal information in a user’s Google Account, data from advertisers that partner with Google, and Google’s inferences about a user’s interests. This last category includes a list of factors, such as apparel or combat sports, that Google’s ad system believes are relevant to the user and thus considers when targeting and delivering ads to them. Users have the option to remove any of these factors from the list of active factors considered by Google’s systems. This page also links users to an install page for a browser plugin that enables users to maintain their preferences for opting out of personalized ads from Google. The plugin enables users to retain these preferences even after they have cleared their cookies. However, these features and controls are only available to Google users who are logged in, leaving non-logged in users or users without a Google account with fewer controls.

According to Google, there were 2.5 billion visits to the Google Accounts page in 2018, but only 20 million people per month visited the ad settings page.⁶³ This raises concerns that although Google offers a limited set of controls to users, users may be either unaware of these options or the settings may be hard to access or understand.

Users and non-users can also adjust ad settings using AdChoices, where they can control ads from other ad networks. This tool provides a list of all of the participating companies including which of them run personalized ads for a user’s browser. The tool also enables users to opt out of data collection and use from participating companies. These participating companies abide by the Digital Advertising Alliance’s (DAA) self-regulatory principles⁶⁴

Additionally, in May 2019, following decisions by browser companies such as Mozilla and BRAVE to restrict the use of cookies, Google implemented similar restrictions on Google Chrome, noting that it wanted to provide users with greater controls over how these cookies were deployed to track their behaviors. The changes included requiring website developers to flag when cookies can be used for cross-site tracking purposes. The changes also allowed users to clear cross-site tracking cookies, while enabling single domain cookies to continue operating, thus maintaining user logins and settings.⁶⁵ Also in May, Google introduced an auto delete feature that complemented the existing manual deletion controls. The new feature enables users to set up automatic deletion of their location, app, and web history. Users are able to choose a period of time—either three months or 18 months—that Google can save their data. Data that falls outside this period is deleted automatically. Users also continue to have the option to pause Google’s web and app activity tracking.⁶⁶

Further, Google introduced a browser extension that users can install to receive more information about the ads they see across Google products and services and from its ad network partners. This information includes the names of third-party companies that helped with the ad targeting process, as well as the names of companies with trackers embedded in ads.⁶⁷

Despite Google’s efforts to promote greater transparency and accountability around its online ad practices, the company falls short. As described, it offers users a limited set of controls over their online ad experience, but these are difficult to understand and use. In addition, non-users or logged out users do not have access to these controls and are therefore unable to manage and opt out of receiving personalized and targeted ads. Further, although Google publishes a political advertising database and transparency report, this report fails to provide granular information on the targeting parameters of these ads as well as related engagement metrics. This report also does not provide this information on an adequate global scale, despite the fact that Google has a global user base. This report also does not provide a comprehensive overview of total advertising on Google’s ad platform, as it only covers political ads. Given that the platform is one of the largest digital advertising players in the industry, it needs to do more to provide adequate and meaningful transparency and accountability around its ad targeting and delivery operations to its users and the public.