This is the fifth installment of a new white paper series on alternative regulatory responses to misinformation. The articles, which were produced with the editing and support of the WIII Initiative, will be presented at a series of events on March 8, 10, and 12th. You can find more information, and registration details for our launch events, here.
While most people might associate fake news with politicians, a substantial component of the current misinformation crisis threatening democracies is websites producing misinformation for commercial reasons. The pandemic has exacerbated this problem, as it provides fertile ground for conspiracy theories and those willing to publish them for ad profit. A responsible press should not indulge their readers’ desire to project fanciful theories onto complex global problems. However, unscrupulous players profit by emulating the communication style of traditional news outlets, but without going through the effort of applying journalistic guidelines to produce accurate information.
The reason this business model is viable is that an increasingly large share of advertising money is allocated by algorithms. Shady publishers would not attract substantial revenue if companies were making individualized choices of where to place their ads. But microtargeting has removed advertisers’ incentive to associate their brand solely with reputable news sources. Instead, companies are served neatly carved demographics, and make decisions based on who to target, rather than which website to sponsor. Automated decisions on ad placement can be obscure even to advertisers themselves. A regulatory response which introduced more transparency to programmatic advertising, thus realigning the incentives of advertisers, would deny ad revenue to harmful misinformation publishers. The best way to achieve this is to provide companies, and the public, with access to large scale, detailed data on advertising spending that is cross-referenced against databases of fact-checked content.
Fact-checking has become popular in recent years as a central journalistic defense against misinformation. While its usage to directly counter fake news remains crucial, combining fact-checks with enhanced advertising transparency can serve to illuminate the money trail which drives the production of misinformation. Transparency could help to increase accountability and oversight around programmatic advertising, to track how resources are channeled to fake news outlets. This paper proposes a framework of enhanced transparency obligations by digital advertising platforms to shed light on the profile, size and impact of ad-subsidized online misinformation. Creating open access repositories of information regarding the resources that fund fact-checked content would increase public knowledge about the problem and its patterns, providing a strong disincentive for companies who currently sponsor misinformation through microtargeted advertising.
Part II of this paper presents the problem: how exploitative misinformation funded by programmatic advertising jeopardizes the health of the public sphere. Part III describes how the platforms’ voluntarily published ad libraries, including lists of political advertisements and basic information about them, are one of the most effective responses so far to this problem. Part IV offers a proposal to mandate an increase in the scope of ad libraries, cross-referencing these libraries with the results of fact-checking processes. Part V concludes.
II. The Problem
Across all constitutional democracies, material published on regular websites and blogs, and funded by advertising in the form of banners, is a crucial piece of the online misinformation puzzle. Financial gain is as much a motivation for fake news production as political gain. For-profit misinformation or exploitative misinformation is particularly detrimental to a healthy public forum because it is produced at a faster pace and disseminated in an organized, profit-driven manner. Indeed, “the primary driver behind the recent explosion of fake news online is advertising income.” This may be distinguished from fake news posted for purely ideological or electoral reasons, even if political actors play an important role in the spread of for-profit fake news produced by third parties. Despite experiments suggesting people are able (to some extent) to identify fake news, they are also more likely to believe misinformation aligned with their political affiliations.
Misinformation which masquerades as legitimate journalism is not a new phenomenon. What has tipped the scales in favor of misinformation in recent years is the new funding system of programmatic advertising, whereby advertisers pick the profile of people who they would like to be exposed to their ad, and algorithms then choose the venue regardless of the integrity of the underlying content. This means that companies no longer consciously choose ad space based on the reputation of a media outlet.
Part of the problem lies in the way that Google and Facebook’s business models interact with one another. User behavior in relation to paid content on Facebook is a determinant of the flow of readers to off-platform misinformation on websites and blogs, which in turn are funded by programmatic ads courtesy of Google. The two companies are deeply intertwined in the ecosystem of fake news and even after the public focus on curbing misinformation since 2016, they continue to exploit it for profit.
As a result, the essential funding infrastructure for media is no longer one of open and accountable decisions regarding ad placement, where companies seek to associate their image with that of a reputable news source. Instead, advertising resources are guided by automated decision-making performed by obscure, platform-driven algorithms. This is precisely the “model required for fake news to flourish”. Algorithms are trained to push engaging news content regardless of its integrity, and in a manner which is not subject to public scrutiny. As hoax websites gradually increase their revenue, swallowing a larger share of a revenue stream that emanates largely from Google Ads, their economic model also competes with legitimate news media, crowding the latter out.
Programmatic advertising is driven by microtargeting, which creates a lack of broader accountability. People know the ad they were shown but have no cost-effective way of learning what ads the digital platforms offered others. Microtargeting also threatens the health of the public discourse by fragmenting the electorate into groups whose understanding of reality may be mutually incompatible. Additionally, microtargeting for political advertising “has a unique risk of harm not associated with traditional political advertising: interference with the rights to privacy and data protection.”
Transparency matters because it affects citizen behavior towards political advertising, and therefore reader attitudes toward publishers that profit from using fake news to tap into political controversies. In an experiment to test the extent to which current ad disclosure features on Facebook were effective at drawing the attention of users, Matthew T. Binford and several coauthors found that the transparency measures implemented by the social media giant are not enough to ensure user awareness. However, if readers know that a message is a paid advertisement, they attribute less credibility to it, share it less often, and also view the brand more negatively.
Much of the current flow of misinformation is supplied by publishers whose incentive structure is opaque. The programmatic advertising system rewards editors whose publishing is built around maximizing attention. Transparency could illuminate the resources financing specific instances of fake news, exposing disingenuous publishers. Over time, this would change reader attitudes towards such publishers and negatively impact the brand image of their advertisers, thus shaming the companies feeding the continuous exploitative misinformation system.
III. Existing Ad-Transparency Initiatives
Ad Libraries as a Driver of Accountability
The most effective tool to support greater transparency in the advertising ecosystem has been the publication of ad libraries. These are repositories curated by digital platforms listing all promoted content, including relevant metadata, such as the money paid to boost the reach of a post and basic information about the people who actually saw it. Facebook and Instagram launched an ad library in 2018, followed by Twitter and then Google, though the latter only includes candidate ads.
There are a number of key benefits flowing from these libraries, particularly the fact that they reveal the resources funneled into political marketing through digital advertising platforms. The libraries uncover the profile and patterns of the non-organic social media operations of political campaigns, allowing the press, academia, and NGOs to digest this information and share their analyses with the public. If a specific post misinforms constituents, there is significant public interest in releasing data about its promotion and targets – data which was wholly unavailable before ad libraries. Ad libraries provide valuable knowledge on the content, authors and audience of artificially spread fake news. They also allow researchers to track the impact of particular campaigns after the fact, by looking for changes in behavior by those exposed to misinformation.
In addition to voluntary efforts by digital platforms, many jurisdictions have pushed for increased transparency in the advertising space by regulating programmatic advertising disclosures, especially connected to elections. The 2018 European Union Code of Practice on Disinformation requires adhering companies to disrupt exploitative misinformation models, restricting advertising revenue through measures which include partnerships with fact-checking organizations. In 2019, the European Commission extolled the ad libraries put up by Facebook, Google and Twitter, but criticized their shortcomings in design (lacking data on issue-based ads) and implementation (technical difficulties).
In the United States, a bill introduced in 2019, named the Honest Ads Act, would require digital platforms under specific circumstances to publish “a public file of all electioneering communications purchased” that would include “a digital copy of the advertisement, a description of the audience the advertisement targets, the number of views generated, the dates and times of publication, the rates charged, and the contact information of the purchaser.” The bill is part of a list of legislation that the 2021 US Congress is considering in order to curtail the digital platforms’ power and lack of accountability.
In Brazil, a draft known as “the fake news bill” was quickly approved in the Senate in 2020 and is under consideration in the House of Deputies as of February 2021. Among a number of reforms, including some very problematic ones, the bill creates two major obligations for social media companies with respect to their advertising disclosures. First, the companies are required to publish an open database of all advertisements which mention a candidate, coalition or political party, regardless of whether these ads take place during a campaign. The database must include the identity of the advertiser, amount spent, and duration of the ad campaign. Second, drawing inspiration from the German Network Enforcement Act, the companies are required to publish quarterly content moderation reports, including figures for the total amount of content published by bots and artificial content distribution networks, the so-called unidentified sponsored and advertising content detected by the company, as well as the corresponding action taken.
Shortcomings of Ad Libraries
Ad libraries, as they currently exist, have a number of shortcomings. First, they are limited in their scope. They only encompass ads by politicians, parties, or their representatives. Selection criteria are limited to election-related advertising, and not all libraries include “issue ads”.
Limiting the ads available in these libraries to those that are classified as “political” fails to provide an adequate solution to track misinformation flows. Even during elections, exploitative misinformation does not originate solely from websites and profiles explicitly linked to candidates and parties. The identification and classification mechanisms used to sort political ads are also prone to error. Using automation to define which ads are political in nature – because their content is political or because the website in which they are displayed is political – creates a severe risk of inaccuracy, particularly when combined with a lack of reliable data about how the algorithms from different companies actually perform. Furthermore, this method of separating political ads from the rest can be more easily gamed.
A second problem is that some libraries only cover advertising spending during election season. Fake news is a challenge to the public discourse year round, as it influences what citizens do not only as voters, but also as communicators who guide the public discourse. Fortunately, approximately a year after their ad library was first created, Facebook decided to increase its scope to encompass any ad run by pages, instead of only those classified as political ads. Twitter went one step further and banned political ads altogether, though it did not provide information on how it will apply this rule to issue ads. Twitter still does not provide data for ads it considers non-political.
A third problem relates to access conditions and usability. The Facebook ad library, for example, is riddled with bugs that make it impossible for information to be extracted. Its requirement for researchers to confirm their identity prior to obtaining access is also problematic. Public interest information should be openly accessible, rather than restricted through demands for personal information and a need to provide statements regarding one’s purpose for access. Companies currently deny access to ad data in bulk, either directly, as Twitter does, or indirectly, as Facebook does by limiting the number of search results in queries. While technical glitches are likely to be fixed eventually, accessibility problems which stem from deliberate design choices are more concerning.
A fourth problem relates to the inconsistent geographic scope of ad libraries. Google provides information for ads based only in Australia, the European Union, the United Kingdom, India, Israel, New Zealand, Taiwan, and the United States. Facebook includes a much larger number of countries from more continents. The Twitter library, discontinued since the social media company banned political ads entirely in 2019, offers data for older political ads that targeted most countries, but the information on issue ads is limited to the United States.
A fifth problem concerns inconsistency in the data catalogued by ad libraries. Given the crucial role that microtargeting plays in the programmatic advertising funding system for misinformation, data on the characteristics of the audience chosen by the advertiser is vital for ad libraries to fulfill their role. Twitter includes all information about how the ad was targeted. Google provides information on all the targeting options available for political ads, namely, geographic location, age and gender. Facebook does not provide any targeting information, only data about the profile of users who actually viewed the ads.
Of these five issues, the thematic scope of ad libraries is the most pressing. By limiting their libraries to “political” or “issue” ads, Google and Twitter deny the public access to vital data about exploitative misinformation that often does not fit neatly into such categories. Fake news is not a problem exclusively about elections or during elections, it is a malady of the networked public sphere that, in turn, hinders the functioning of democracy. Coronavirus hoax messages and websites, for example, have proven to be a nightmare for public health officials in 2020, with knock-on effects in the political realm.
Furthermore, misinformation is not produced and spread solely by politicians or parties. Much of the content published by hoax websites has no direct connection to specific politicians. Antivaxx misinformation impairing COVID-19 inoculation initiatives is both a more pressing problem than libel of candidates for office, and one that runs much deeper.
A significant amount of the money which drives misinformation is absent from the Google and Twitter ad libraries because of the design decision to limit transparency to “political” or “issue” ads. This prevents civil society from accessing and analyzing valuable data about how fake news is promoted, or scrutinizing the broader role of advertising in the misinformation ecosystem. Likewise, the Facebook ad library amasses all ads together without classification options that would make it possible to isolate fake news ads. The following section lays out a proposal for remedying these deficiencies, by mandating that ad libraries should be raised to a higher common standard, to facilitate better oversight of the misinformation ecosystem.
IV. The Proposed Response
This paper proposes that that all major platforms be mandated to set up ad libraries of all paid or sponsored content, populated according to universal standards, including metadata regarding the URL, content, identity of the advertiser, amount spent, and all audience profile targeting options selected by the advertiser. The set of data about each ad and the format accessibility requirements in the proposal create obligations that go beyond current voluntary ad libraries. Such increased transparency might not be necessary in all cases, but this proposal targets a specific group of advertised content with a strong indication of abuse: content that has been fact-checked by accredited fact-checking organizations. The libraries should be designed with an eye towards ease of use, particularly by sophisticated users, including ensuring that the material is published in open and machine-readable formats.
The metadata about each ad should include, at a minimum, the URL, content, identity of the advertiser, amount spent, and all audience profile targeting options selected by the advertiser. The full repository should be available for download, and ideally there should also be an option to select data according to a particular timeframe. Access should not be curtailed by offering only the results of limited searches, or the visualizations presented in interactive dashboards. These features are useful to certain categories of readers, but they should be offered as an alternative to downloading the raw data. It is crucial for experienced users to be able to easily obtain the entirety of the relevant data they are looking for.
The most novel element of this proposal is the criterion to allow for the ad library to be cross-referenced against the results of fact-checks. In order to serve as a robust tool against misinformation, ad libraries should be sorted according to “assessments of the validity of claims made by public officials and institutions with an explicit attempt to identify whether a claim is factual”.
To be clear, the point is not that all promoted content should be subsequently fact-checked. That would be prohibitively costly and overbroad. Rather, platforms are to cross-reference their databases of promoted content, on one side, and databases of fact-checked content, on the other, such that it is possible to produce a new open dataset containing all promoted content that has been fact-checked. While there is some cost to sorting promoted content in this manner, this is a reasonable burden for the digital platforms to bear, given their own role in the problem.
The point of this proposal is less about increasing the chance that people exposed to false ads are also inoculated with a fact-check, than it is about providing increased public scrutiny over the misinformation ecosystem. Nonetheless, the tested effects of fact-checking misinformation do matter. Even before 2016, scientific research already pointed to fact-checking’s ability to sway the opinion of the reader. Politically-savvier people have an increased chance of being influenced by fact-checking results, though fact-checking initiatives can also resonate with less politically sophisticated readers. Indeed, fact-checking results can “positively affect beliefs, irrespective of political ideology, preexisting positions, context (campaign vs. routine), and whether it refutes the entire false statement or just parts of a statement”, though this is less effective during elections.
Just as algorithmic moderation obfuscates what news stories are actually shown on recommended feeds and popularly read, frustrating accountability for social media platforms, programmatic advertising hides where a company has placed an ad, which prevents public accountability against both the publisher and the advertiser. Boosting the clarity and accessibility of ad libraries would not only convey useful information to the public at large, but also to the companies paying for the ads in the first place, giving them back awareness of where their ads appear, and thus some of the control that programmatic advertising has taken away.
Sleeping Giants, a global activism organization which works to pressure advertisers who support problematic content, currently carries out an effective strategy based on manually taking screenshots of ads displayed on particular websites. With these enhanced ad libraries, digital platforms would publish this material automatically, allowing campaigns against fact-checking misinformation to be effectively scaled. Civil society organizations, academia, and the press would have access to much more relevant material to engage with the public discourse around misinformation. To be sure, “activism around programmatic advertising is necessarily complex and therefore challenging”. However, public perceptions towards the brands that inadvertently fund misinformation are a key piece of the puzzle, as Sleeping Giants’ work shows. Evidence demonstrates that a connection with fake news adversely impacts the public image of companies, by causing the reader to discredit the outlet. Lower levels of source credibility in turn negatively impact the reader’s opinion of the advertised brand. A “new consumer movement is rising” around this exposure dynamic, providing traction to traditional regulatory tools such as transparency, in combination with market effects such as brand perception and corporate reputation management.
This proposal is a variation of an initiative that Facebook, Google and Twitter have already taken on, which suggests that it should not be excessively costly or challenging to implement. As a regulatory response, it does not deny consumers any service or feature, and as such does not stifle innovation. Facebook, Google and Twitter each offer some of the components described above, but never all of them together. However, while ad libraries already exist in various forms, a comprehensive version, combined with international fact-checking databases, would be a major step forward in terms of supporting critical work by civil society and journalists in shedding light on how misinformation spreads. It should be clear, however, that this proposal does not in any way entail information currently published in ad libraries to be removed if it does not relate to any fact-checking. This is about mandating the publication of additional data in a specific format, not about erasing any information already in the public realm.
Abby Wood and Ann Ravel have argued that the availability of open ad lists would enable citizens to know when they are targeted by “divide and conquer”-style exploitative misinformation schemes, perpetrated through programmatic advertisements, while also preventing ads from being used to squash counter speech. Likewise, enhancing transparency in the advertising space would support projects like NewsGuard, which works “with the advertising industry to make apolitical, transparent trust ratings for news sites available to advertisers on large ad exchanges”. One of the key challenges to addressing exploitative misinformation is that we do not know what we do not know. Ideally, empirical scholarship on platform practices should inform smart regulation. As it stands, the law often empowers companies to prosecute researchers for the retrieval and use of publicly available data on their digital platforms.
The most technically challenging aspect of this proposal is that platforms would have to decide whether the vast the amounts of promoted content which they host have been fact-checked. While it is easy enough to create a list of URLs which have been specifically subject to fact-checks, carrying out an analysis as to whether this content appears elsewhere is more challenging. Carrying out this task with regard to videos or images adds yet another layer of complexity.
However, in considering the technical challenges underlying this problem, it is worth bearing in mind that digital platforms have years of experience in the automated identification of specific types of content. Implementing the proposal, either in compliance with legislation or on a voluntary basis, would require an organizing effort that is automatic, transparent, ex post and centralized. Artificial intelligence is already a common tool in content moderation, and techniques such as the development of adversarial networks have greatly increased the accuracy of automated decisions. Google, for instance, notoriously uses AI in its Content ID feature to make moderation decisions. While there are significant criticisms with how Content ID has been implemented to remove user-generated content, these concerns do not apply in the same way with regards to automated decisions for ad libraries. While false positives generated under this proposal are concerning insofar as they could undermine the efficacy of the ad library as a tool for transparency, they do not raise the same freedom of expression concerns as the existing use of Content ID.
Another more complex element is deciding what repositories of fact-checks to use. If the platforms refused to implement this proposal voluntarily, devising a regulatory structure which defined the scope of an “acceptable” fact-check could be both technically challenging and, in some instances, engage difficult constitutional questions. Any regulatory requirement would have to be precise enough to enable enforcement, but the task of defining the scope of the repository would ideally be delegated to an independent agency. There are numerous fact-checking agencies operating in different countries, under varying institutional governance systems, such as the “newsroom model” and the “NGO model”, with varying levels of source transparency.
One starting point could be to defer to the International Fact-Checking Network, and in particular to reference the signatories to their code of principles. The list currently has 29 members, including in South Africa, France, Spain, the United States, Brazil, Serbia, Turkey, and Bosnia & Herzegovina. The Reporter’s Lab at Duke University manages a separate database of over 300 active fact-checking institutions worldwide that have met certain criteria, including transparency of sources, disclosure of affiliations, and an even-handed approach to checking statements from different political parties. These lists could be a starting point for determining the parameters of the ad libraries.
Programmatic advertising has both removed the transparency of ad placement and restricted advertisers’ choices in deciding who to support. As microtargeting selects based on a specific set of eyeballs instead of the location of ads, publisher reputation is no longer an element that directly influences which ad is shown where, with a detrimental impact on the commercial value of journalistic standards. This has drained resources from reputable news media, directing a continuous flow of opaque funds to hoax websites.
Transparency constitutes a structure-oriented measure for fighting misinformation, whereby regulation creates increased disclosure obligations that uncover the money trail of for-profit fake news. Instead of command-and-control regulatory responses that invite more platform censorship in order to fight abusive content while denying civil society a meaningful role, this is a solution that looks to transparency obligations as a procedural guarantee. In recent years, digital platforms have created repositories of political ads to try and remedy the problem. However, these initiatives are fundamentally flawed in that their visibility criteria are skewed towards election-related advertising, and they fail to comprehensively illuminate the funding which drives the production of misinformation.
Platforms should include information in their ad libraries about the promotion and financial subsidy of any content that has been fact-checked, regardless of whether these companies believe it to be “political”. This is not a revolutionary measure that would entail major new costs, but rather a logical next step in a policy that these digital platforms have already acknowledged as warranted and started to gradually implement.
In implementing this regulatory response, governments should recognize the central role played by fact-checking agencies in a healthy public sphere and avoid much harsher, complex and speech-restrictive measures that would target misinformation content for removal by granting digital platforms undue power and responsibilities over the shape of online speech. Enhanced ad libraries are a natural response as they fix an obscurity problem with disclosure, realigning advertiser and publisher incentives without trending towards problematic forms of censorship.
Ivar A. Hartmann is an Associate Professor at the Insper Learning Institution, Brazil.
 Hunt Allcott & Matthew Gentzkow, Social Media and Fake News in the 2016 Election, 31(2) Journal of Economic Perspectives, 211 (2017) at 218.
 David M. Lazer et al., The science of fake news, 359 Science 1094 (2018) at 1094
 Hunt Allcott & Matthew Gentzkow, supra note 1, at 213.
 Commentary surrounding the German Network Enforcement Act (NetzDG) aptly illustrates the chilling effects of command-and-control regulation of online speech and its disproportionate restriction of the right to freedom of expression. See, e.g., Gerald Spindler, Internet Intermediary Liability Reloaded The New German Act on Responsibility of Social Networks and its (In-) Compatibility with European Law, 8 JIPITEC 166 (2017). On the other hand, the section of the NetzDG creating transparency obligations for social media regarding their moderation practices is in line with the broader focus of transparency in this proposal. So far, it seems the main obstacle in its enforcement is companies directing users to register complaints with grounds different from those in the law, such that a large share of moderation decisions fall outside of the scope of the transparency reports. Ben Wagner et al., Regulating Transparency? Facebook, Twitter and the German Network Enforcement Act, in FAT* '20: Conference on Fairness, Accountability, and Transparency (2020), https://dl.acm.org/doi/abs/10.1145/3351095.3372856 (last visited Feb 1, 2021).
 Richard Graham, Google and advertising: digital capitalism in the context of Post-Fordism, the reification of language, and the rise of fake News, 3(45) Palgrave Communications 1 (2017), at 12.
 Giandomenico Di Domenico et al., Fake News, social media and marketing: A systematic review, 124 Journal of Business Research 329 (2021) at 337. There is also a third motivation, based on more anarchic or trollish instincts, and which is more resistant to traditional fact-checking. See Brooke Borel, Fact-Checking Won’t Save Us From Fake News, FiveThirtyEight (Jan 4, 2017), https://fivethirtyeight.com/features/fact-checking-wont-save-us-from-fake-news/.
 Adam J. Mills, Christine Pitt & Sarah Lord Ferguson, The Relationship between Fake News And Advertising Brand Management in the Era Of Programmatic Advertising and Prolific Falsehood, 59(1) Journal of Advertising Research 3 (2019), at 4.
 “Carroll estimated that a fake-news share from within the Trump campaign could earn the lucky hoaxer as much as $10,000 in extra revenue, provided they have taken full advantage of the ad services available to them. That’s a “huge economic incentive to create stories that they want to distribute.” Abby Ohlheiser, This is how Facebook’s fake-news writers make money, The Washington Post (Nov 18, 2016), https://www.washingtonpost.com/news/the-intersect/wp/2016/11/18/this-is-....
 Andrew Guess et al., Fake news, Facebook ads, and misperceptions. Assessing information quality in the 2018 U.S. midterm election campaign (2019), http://www.dartmouth.edu/~nyhan/fake-news-2018.pdf, at 21.
 “When brands can track desirable users across the web, they not only have less incentive now than they once did to think about editorial context, they may see a benefit in reaching those users in the cheapest possible spaces—spaces that tend not to belong to reputable publishers, but to clickbait artists.” Joshua A. Braun, John D. Coakley & Emily West, Activism, Advertising, and Far-Right Media: The Case of Sleeping Giants, 7(4) Media and Communication 68 (2019), at 70.
 “Combined with the media- specific characteristics of search engine results and social media feeds that decontextualize individual articles and present a diverse range of content, Google and Facebook encourage a logic that incentives clickbait headlines. (…) As long as profits are tied directly to how much an article is shared or viewed then very particular kinds of media content will continue to be incentivized over others. Fake news is just one example to consider when investigating how Google creates avenues for profit and how Google’s economics co-depend on other online institutions, in particular, Facebook.” Richard Graham, supra note 5 at 17.
 Emily Bell & Taylor Owen, The Platform Press: How Silicon Valley reengineered journalism, Tow Center for Digital Journalism (Mar 29, 2017), https://www.cjr.org/tow_center_reports/platform-press-how-silicon-valley-reengineered-journalism.php. The story is now famous of fake news producers from Macedonia who earned a proportionally tremendous sum of money: “He claims to have earned up to $2,500 a day from advertising on his website, while the average monthly income in Macedonia is just $426.” Florence Davey-Attlee & Isa Soares, The Fake News Machine: Inside a Town Gearing up for 2020, CNN (Sept 13, 2017), https://money.cnn.com/interactive/media/the-macedonia-story/.
 Adam J. Mills, Christine Pitt & Sarah Lord Ferguson, supra 7 note at 5. See also Richard Graham supra note 5, at 17.
 Jeremy B. Merrill & Marshall Allen, “Trumpcare” Does Not Exist. Nevertheless Facebook and Google Cash In on Misleading Ads for “Garbage” Health Insurance, ProPublica (Oct 20, 2020), https://www.propublica.org/article/trumpcare-does-not-exist-nevertheless-facebook-and-google-cash-in-on-misleading-ads-for-garbage-health-insurance.
 Richard Graham, supra note 5 at 14.
 Joshua A. Braun & Jessica L. Eklund, Fake News, Real Money: Ad Tech Platforms, Profit-Driven Hoaxes, and the Business of Journalism, 7(1) Digital Journalism 1 (2019), at 19.
 Andrew Guess et al., supra note 9, at 2.
 Abby K. Wood & Ann M. Ravel, Fool Me Once: Regulating Fake News and Other Online Advertising, 91 S. Cal. L. Rev. 1223 (2018), at 1236.
 Tom Dobber, Ronan Ó Fathaigh & Frederik J. Zuiderveen Borgesius, The regulation of online political micro-targeting in Europe, 8(4) Internet Policy Review 1 (2019), at 13.
 Matthew T. Binford et al., Invisible transparency: Visual attention to disclosures and source recognition in Facebook political advertising, Journal of Information Technology & Politics 1 (2020, ahead of print) at 11. A lesser-known politician in the ad seems to render people unable to remember it was even an ad in the first place, meaning that the risk of explicitly marked ads not being recognized as such is high.
 Id., at 4.
 Juan Carlos Medina Serrano et al., Exploring Political Ad Libraries for Online Advertising Transparency: Lessons from Germany and the 2019 European Elections, in SMSociety'20: International Conference on Social Media and Society (2020), https://dl.acm.org/doi/abs/10.1145/3400806.3400820 (last visited Feb 2, 2021), at 111.
 “The labeling won’t just apply to candidate and election ads, but those dealing with political issues such as “abortion, guns, immigration or foreign policy.” John Constantine, Facebook and Instagram launch US political ad labeling and archive, TechCrunch (May 24, 2018), https://techcrunch.com/2018/05/24/facebook-political-ad-archive/. The library (initially named Ad Archive) is available at https://www.facebook.com/ads/library/ (last visited Feb 2, 2021).
 Anthony Ha, Twitter launches its Ads Transparency Center, where you can see ads bought by any account, TechCrunch (June 28, 2018), https://techcrunch.com/2018/06/28/twitter-ads-transparency-center/. The library is available at https://business.twitter.com/en/help/ads-policies/product-policies/ads-transparency.html (last visited Feb 2, 2021)
 Taylor Hatmaker, Google releases a searchable database of US political ads, TechCrunch (Aug 15, 2018), https://techcrunch.com/2018/08/15/google-political-ad-library/. The library is available at https://transparencyreport.google.com/political-ads/home (last visited Feb 2, 2021).
 “Relevant Signatories commit to deploy policies and processes to disrupt advertising and monetization incentives for relevant behaviors, such as misrepresenting material information about oneself or the purpose of one’s properties. These policies and processes can include, for example, the restriction of advertising services or limiting paid placements, and could potentially take place in partnership with fact-checking organizations.” European Union, EU Code of Practice on Disinformation (2018), https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=54454 (last visited Feb 3, 2021), at 5.
 “In particular, Facebook and Twitter have made political advertisement libraries publicly accessible, while Google's library has entered a testing phase. This provides the public with more transparency around political ads. However, further technical improvements as well as sharing of methodology and data sets for fake accounts are necessary to allow third-party experts, fact-checkers and researchers to carry out independent evaluation. At the same time, it is regrettable that Google and Twitter have not yet reported further progress regarding transparency of issue-based advertising, meaning issues that are sources of important debate during elections.” European Commission, Code of practice against disinformation: Commission welcomes the commitment of online platforms ahead of the European elections, European Commission (April 23, 2019), https://ec.europa.eu/commission/presscorner/detail/en/STATEMENT_19_2174.
 Senator Mark R. Warner, The Honest Ads Act, https://www.warner.senate.gov/public/index.cfm/the-honest-ads-act (last visited Feb 3, 2021).
 Lauren Feiner, Democratic Senate control could breathe new life into net neutrality and other tech regulations, CNBC (Jan 9, 2021), https://www.cnbc.com/2021/01/09/democratic-senate-control-could-breathe-new-life-into-tech-policy.html.
 Previous versions of the bill tried to produce a viewpoint specific concept of fake news that would invite State censorship while also creating exacting requirements for people subscribing to social networks. See Raphael Tsavkko Garcia, Brazil’s “fake news” bill won’t solve its misinformation problem, MIT Technology Review (Sept 10, 2020), https://www.technologyreview.com/2020/09/10/1008254/brazil-fake-news-bill-misinformation-opinion/. The draft eventually approved by the Senate was “considerably dehydrated” but still contained a few controversial elements. Angelica Mari, Brazilian Senate passes fake news bill, ZDNet (July 1, 2020), https://www.zdnet.com/article/brazilian-senate-passes-fake-news-bill/.
 See articles 13, V and 15. Senado Federal do Brasil, Projeto de Lei n° 2630, de 2020, https://www25.senado.leg.br/web/atividade/materias/-/materia/141944. The Center for Technology and Society at FGV Law School in Rio de Janeiro published a translation of the entire bill, available at https://bit.ly/br_fakenewsbill (last visited Feb 3, 2021).
 “[T]he drawback of ‘issue ads’ as a scoping device, is that the concept of a political ‘issue’ is broad and subjective, and makes it difficult for archive operators to develop actionable definitions and enforce these in practice.” Paddy Leerssen et al., Platform ad archives: promises and pitfalls, 8(4) Internet Policy Review 1 (2019), at 8.
 The press was able to prove certain political ads get through on Facebook without proper listing in the ad library. See Jennifer Baker, Facebook Ad Library: Only 3 ‘Brexit’ ads for whole month. Yeah, right!, TNW (Nov 1, 2018), https://thenextweb.com/contributors/2018/11/01/facebook-ad-library-only-3-brexit-ads-for-whole-month-yeah-right/.
 Juan Carlos Medina Serrano et al., supra 23 note at 119.
 Andrew Hutchinson, Facebook Expands Ad Archive to All Ads and Pages, SocialMediaToday (March 29, 2019), https://www.socialmediatoday.com/news/facebook-expands-ad-archive-to-all-ads-and-pages/551584/. The problematic “political” criterion made a comeback, though, as the platform moved to ban all political ads the week before election in 2020. In addition to false negatives, the policy is ineffective because it overlooks posts that were already being promoted before the final week. Davey Alba & Sheera Frenkel, Why Facebook’s Blocking of New Political Ads May Fall Short, The New York Times (Sept 4, 2020), https://www.nytimes.com/2020/09/04/technology/facebooks-political-ads-block-election.html.
 Amrita Khalid, Twitter’s political ad ban won’t stop politicians getting their messages out, Quartz (Nov 6, 2019), https://qz.com/1742852/campaigns-arent-worried-about-twitters-political-ad-ban/.
 These technical issues were uncovered by researchers from two separate initiatives. “In general, we encountered three categories of issues with the Facebook Ad Library API. First, software programming errors that cripple a user's ability to complete even a single search (…) Second, technical or data issues that affect a user's ability to reliably retrieve data from multiple searches”. Mozilla Foundation, Data Collection Log — EU Ad Transparency Report, Mozilla Foundation (2019), https://adtransparency.mozilla.org/eu/log/ (last visited Feb 2, 2021). See also French Ambassador for Digital Affairs, Facebook Ads Library Assessment, French Ambassador for Digital Affairs (2019), https://disinfo.quaidorsay.fr/en/facebook-ads-library-assessment (last visited Feb 2, 2021).
 “We’ll need to confirm your identity and location. If you haven't already confirmed your ID, it typically takes 1-2 days to complete this step.” Facebook Help Center, What is the Facebook Ad Library and how do I search it?, Facebook, https://www.facebook.com/help/259468828226154 (last visited Feb 2, 2021).
 “The real genius of FOIA, however, is its provision permitting anyone to seek access to government-held information upon request. (…) It allows "any person" -including corporations, nonprofit entities, and even foreign nations -to request any record from any federal agency or government-controlled entity on any subject, without saying why the record was requested or what purpose disclosure would serve.” David C. Vladeck, Information Access - Surveying the Current Legal Landscape of Federal Right-to-Know Laws, 86 Tex. L. Rev. 1787 (2008), at 1797.
 Matthew Rosenberg, Ad Tool Facebook Built to Fight Disinformation Doesn’t Work as Advertised, The New York Times (July 25, 2019), https://www.nytimes.com/2019/07/25/technology/facebook-ad-library.html.
 “The USA, for instance, counts 3.8 million ads and we observed that requests cannot ask for more that 2000 ads each. To reach the last page, one has to successfully execute about 1900 requests in order, which we found impossible to achieve in the two weeks we tried.” French Ambassador for Digital Affairs, supra note 38.
 Though the dropdown in the report tool page lists 83 countries, the source code for the page includes a much larger number, perhaps as a sign of plans to increase to more countries in the future. https://www.facebook.com/ads/library/report/ (last visited Feb 2, 2021).
 Juan Carlos Medina Serrano et al., supra note 23 at 118.
 “The information imbalance between a voter who has little time to inform herself on politics and the political or media speaker is often profound. This imbalance gives the voter reason to trust the veracity of a piece of false campaign speech.” Alvin I. Goldman & Daniel Baker, Free Speech, Fake News, and Democracy, 18 First Amendment Law Review 66 (2019), at 129.
 Jesselyn Cook, A Toxic ‘Infodemic’: The Viral Spread Of COVID-19 Conspiracy Theories, HuffPost (July 4, 2020), https://www.huffpost.com/entry/coronavirus-conspiracy-theories-social-media_n_5e83d701c5b6a1bb764f6d3b. Since this story was published, Buttar’s channel on YouTube has reached 500.000 subscribers: https://www.youtube.com/user/drbuttar (last visited Feb 2, 2021).
 Political parties are responsible for 86% of ads, but only 48% of the spending. See Juan Carlos Medina Serrano et al., supra note 23 at 118.
 The testimony of a nurse from Arizona is compelling: “‘We are feeling like our community — at least a good section of them — have turned on us,’ he said. ‘It’s just amazing that nurses have gone from the number one trusted profession in the United States to now realizing that there’s this subsection of the population that thinks we’re trying to kill everyone.’” Jane Lytvynenko, In 2020, Disinformation Broke The US, BuzzFeed News (Dec 6, 2020), https://www.buzzfeednews.com/article/janelytvynenko/disinformation-broke-us.
 The proposal is mainly directed at Facebook and Google, for the reasons already laid out in previous sections, though other large digital platforms with significant advertising revenue arguably should also be subjected to the same rules. “DPs are quickly controlling news distribution. Facebook is now the second largest news provider in terms of attention share in the United States. In the UK, Facebook is third, Google is fifth, and Twitter is tenth. By curating the news viewers receive, DPs have effectively appropriated the role that newspaper editors used to have in influencing readers’ attention. This poses a concentration issue, as thousands of different viewpoints have now been replaced by a duopoly.” Luigi Zingales & Filippo Maria Lancieri, Stigler Committee on Digital Platforms: Policy Brief, George J. Stigler Center for the Study of the Economy and the State (Sep 2019), https://www.chicagobooth.edu/-/media/research/stigler/pdfs/policy-brief---digital-platforms---stigler-center.pdf, at 5.
 Nathan Walter et al., Fact-Checking: A Meta-Analysis of What Works and for Whom, 37(3) Political Communication 1 (2019), at 2.
 Kim Fridkin, Patrick J. Kenney & Amanda Wintersieck, Liar, Liar, Pants on Fire: How Fact-Checking Influences Citizens’ Reactions to Negative Advertising, 32(1) Political Communication 127 (2015), at 139.
 Id at 145.
 Nathan Walter et al., supra note 50 at 17-18.
 Zeynep Tufekci, Algorithmic Harms Beyond Facebook and Google: Emergent Challenges of Computational Agency, 13 Colorado Technology Law Review 203 (2015).
 Joshua A. Braun & Jessica L. Eklund, supra note 17 at 11. See also Adam J. Mills, Christine Pitt & Sarah Lord Ferguson, supra note 7 at 5: “Programmatic advertising is highly efficient, extremely cost-effective, virtually immediate, and designed almost completely around reaching the individual target consumer. To target in this manner, however, prioritizing consumers over con-text, advertisers must cede almost all control over advertisement placement to the intermediaries.”
 Joshua A. Braun, John D. Coakley & Emily West, supra note 11 at 70.
 “In this vein, researchers should go beyond merely investigating the direct effect of the ad and address the broader set of causal relationships involved in the formation of brand attitude and purchase intentions.” Marco Visentin, Gabriele Pizzi & Marco Pichierri, Fake News, Real Problems for Brands: The Impact of Content Truthfulness and Source Credibility on consumers' Behavioral Intentions toward the Advertised Brands, 45 Journal of Interactive Marketing 99 (2019), at 100.
 Id at 106.
 “(…) a new consumer movement is rising, and activists believe that where votes failed, wallets may prevail. This struggle is about much more than ads on Breitbart News — it’s about using corporations as shields to protect vulnerable people from bullying and hate crimes.” Pagan Kennedy, How to Destroy the Business Model of Breitbart and Fake News, The New York Times (Jan 7, 2017), https://www.nytimes.com/2017/01/07/opinion/sunday/how-to-destroy-the-business-model-of-breitbart-and-fake-news.html.
 Facebook went so far as boasting the ad library to be one of its main measures against misinformation. See Sheryl Sandberg, Hearing Before the United States Senate Select Committee on Intelligence, United States Senate (Sep 5, 2018), https://www.intelligence.senate.gov/sites/default/files/documents/os-ssandberg-090518.pdf, at 5.
 Mark D. Fenwick, Wulf A. Kaal & Erik P.M. Vermeulen, Regulation Tomorrow: What Happens When Technology Is Faster than the Law? 6(3) American University Business Law Review 561 (2017), at 587.
 Abby K. Wood & Ann M. Ravel, supra note 19 at 1259-1260.
 Matt Skibinski, Misinformation won’t stop unless we stop it (Predictions for Journalism 2021, NiemanLab, 2020), https://www.niemanlab.org/2020/12/misinformation-wont-stop-unless-we-stop-it/
 “If we do not know whether social media has overall positive or negative effects on our polity, we have to blame the DPs themselves. All the data they generate is proprietary and they deny outside, independent researchers access to almost all of it. In doing so, they also prevent our societies from comprehending their true impacts.” Luigi Zingales & Filippo Maria Lancieri, supra note 49 at 10.
 Mandating the publication of expanded ad libraries would solve, for the purposes of research about exploitative misinformation, the problem aptly spotlighted by Thomas E. Kadri, Platforms as Blackacres, 68 UCLA L. Rev. 1 (forthcoming 2021), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3742086.
 James Grimmelmann, The Virtues of Moderation, 17(1) Yale Journal of Law and Technology 42 (2015).
 Cambridge Consultants, Use of AI in Online Content Moderation, OFCOM (2019), https://www.ofcom.org.uk/__data/assets/pdf_file/0028/157249/cambridge-consultants-ai-content-moderation.pdf, at 74.
 See Maayan Perel & Niva Elkin-Koren, Accountability in Algorithmic Copyright Enforcement, 19 Stan. Tech. L. Rev. 473 (2016).
 Fact-checking agencies are more likely to be found in countries with “increased democratic governance and Internet accessibility”. Michelle A Amazeen, Journalistic interventions: The structural factors affecting the global emergence of fact-checking, 21(1) Journalism 95 (2020), at 103.
 See Edda Humprecht, How Do They Debunk “Fake News”? A Cross- National Comparison of Transparency in Fact-checks, 8(3) Digital Journalism 310 (2020).
 International Fact-Checking Network, International Fact-Checking Network fact-checkers’ code of principles, https://www.poynter.org/ifcn-fact-checkers-code-of-principles/ (last visited Feb 3, 2021).
 Bill Adair & Mark Stencel, How We Identify Fact-Checkers, Reporter’s Lab (June 22, 2016), https://reporterslab.org/how-we-identify-fact-checkers/.
 Robert Gorwa, The platform governance triangle: conceptualising the informal regulation of online content, 8(2) INTERNET POLICY REVIEW 1 (2019), at 14.
 Barbora Bukovská, The European Commission’s Code of Conduct for Countering Illegal Hate Speech Online. An analysis of freedom of expression implications (The Transatlantic Working Group Papers Series, May 7th, 2019), https://cdn.annenbergpublicpolicycenter.org/wp-content/uploads/2020/05/E..., at 6.
 Ivar Hartmann, A new framework for online content moderation, 36 Computer Law & Security Review 1 (2020).