Political Communication
Q Routledge
Taylor &. Francis Groi
ISSN: 1058-4609 (Print) 1091-7675 (Online) Journal homepage: https://www.tandfonline.com/loi/upcp20
The "Arbiters of What Our Voters See": Facebook and Google's Struggle with Policy, Process, and Enforcement around Political Advertising
Daniel Kreiss & Shannon C. Mcgregor
To cite this article: Daniel Kreiss & Shannon C. Mcgregor (2019): The "Arbiters of What Our Voters See": Facebook and Google's Struggle with Policy, Process, and Enforcement around Political Advertising, Political Communication, DOI: 10.1080/10584609.2019.1619639
To link to this article: https://doi.Org/10.1080/10584609.2019.1619639
Published online: 19Jun 2019.
Submit your article to this journal C?
View Crossmark data C?
Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journal lnformation?journalCode=upcp20
Political Communication, 0:1-24, 2019 Copyright © 2019 Taylor & Francis Group, LLC ISSN: 1058-4609 print / 1091-7675 online DOI: https://doi.org/10.1080/10584609.2019.1619639
|3 Routledge
Taylor &. Francis Gi
I roup
N Check for updates
The "Arbiters of What Our Voters See": Facebook and Google's Struggle with Policy, Process, and Enforcement around Political Advertising
DANIEL KREISS and SHANNON C. McGREGOR
The question of how Facebook and Google make and justify decisions regarding permissible political advertising on their platforms is increasingly important. In this paper, we focus on the U.S. case and presentfindings from interviews with 17former social media firm employees (n = 7) and political practitioners (n = 11). We also analyze emails (n = 45) exchanged between Facebook government and elections staffers and two campaigns, a U.S. gubernatorial (2017) and presidential campaign (2016), regarding the platform's policies in the context of paid speech. In addressing questions about Facebook's and Google s processes and policies regarding paid political content, the rationales for them, and the ability of campaigns to contest decisions, this study shows how while Facebook and Google resist being arbiters of political discourse, they actively vet paid content on their platforms. These platforms differ with respect to how and what decisions they make in the context of paid speech and within each company there are active and ongoing debates among staffers about speech. These debates at times take place in consultation with political practitioners and often occur in the context of external events. Across these firms, policies regarding speech evolve through these internal debates, appeals by practitioners, and outside pressure. At the same time, both Facebook and Google make decisions in often opaque ways, according to policies that are not transparent, and without clear justifications to campaigns or the public as to how they are applied or enforced. This limits options for political practitioners to contest regulation decisions. Finally, we conclude by arguing for the need for expanded capacities for political practitioners and the public to exercise voice around the content decisions that these firms make, and for firms to create more robust institutional mechanisms for incorporating it.
Keywords   Facebook, Google, electoral politics, political advertising, U.S. politics
Over the past three U.S. presidential cycles, digital political advertising and paid content promotion have become central to the efforts of candidates and campaigns to get their messages in front of voters. While there are thousands of ways to purchase and target online ads, Facebook and Google dominate the market for digital advertising in all its forms, together making up 58% of the $111 billion market (Soper, 2018). This has
Daniel Kreiss is Associate Professor in the School of Media and Journalism at the University of North Carolina at Chapel Hill. Shannon C. McGregor is an Assistant Professor in the Department of Journalism at the University of Utah.
Address correspondence to Daniel Kreiss, UNC School of Media and Journalism, Carroll Hall, CB 3365, Chapel Hill, NC 27599, USA. E-mail: dkreiss@email.unc.edu
1
2
Daniel Kreiss and Shannon C. Mcgregor
significant implications for politics. In line with the commercial sector, Facebook and Google are increasingly the platforms that candidates are reliant upon for their strategic digital communications, and especially paid media.
As such, it is crucial for political communication scholars to understand how these platform companies engage in the private regulation of paid political speech. There is a research literature on the regulation of political advertising delivered over broadcast and cable networks, which historically has been the subject of regulation in the United States by both the government (in terms of compelling transparency and disclosure) and these networks themselves (in terms of setting content guidelines) (e.g. Newell & Blevins, 2018). Little is known, however, about the private regulation of paid political speech on digital platforms outside of broad Federal Election Commission guidelines related to disclosure and recent journalistic attention to new political ad transparency initiatives.
This paper offers a first analysis of the paid content moderation policies at Facebook and Google in the context of politics. We focus on paid political advertising because this is the domain that should have the most explicit policies, transparent decision-making, and consistent enforcement, as opposed to content moderation more broadly (see Gillespie, 2018). Unlike organic content created and shared by users, paid political advertising is an explicitly commercial transaction, advertising has always been directly facilitated and requires express advance approval by these firms, and digital political advertising is subject to oversight in the U.S. by the Federal Election Commission. As such, paid political advertising is an important test case for these firms' editorial roles visa-vis political speech more broadly. To that end, in this study we examine Facebook's and Google's processes and policies governing paid political content, the rationale these firms provide for the actions - and inaction - they take in regulating paid political content, and the recourse available to campaigns to contest regulating decisions.
Through interviews with political practitioners and former employees at Facebook and Google, as well as analysis of email exchanges between Facebook employees and staffers on a U.S. gubernatorial and presidential campaign, this paper shows how through policies about acceptable advertising content and an active advertising approval process, Facebook and Google exercise considerable discretion over paid political media. While these decisions are clearly difficult, they are made without much in the way of transparency, consistency, or accountability. And, we show that decisions regarding political advertising content are at times made through human negotiations over the permissibility of political advertising that take place between political practitioners and employees of Facebook and Google, as well as between these employees themselves. This looks different from both the automated and human-vetted forms of content moderation for inappropriate content that Gillespie (2018) details in the context of non-paid user posts.
This paper proceeds in four parts. First, we discuss the relationship between technology firms, their platforms, and political speech. Then, we detail our methods for this study and results. Drawing on interviews with former employees of Google and political practitioners, we first provide an analysis of Google's explicit political advertising policies and how they are implemented and enforced in practice, while also contrasting these policies and procedures with those at Facebook. We then analyze emails between campaign staffers and Facebook employees to show how the firm justifies its actions and inaction, or more often struggles to, in the context of editorial decision-making around paid political content. We conclude with a discussion of the structural tensions between these firms' roles as commercial and political marketplaces and consider how these firms should approach their editorial obligations.
Arbiters of What Voters See
3
Technology Firms and Political Speech
Over the past five years, there has been growing scholarly interest in the workings of technology companies and their social media platforms. This literature is rapidly advancing, but generally encompasses the economic and other relationships between technology companies and firms in other fields such as publishers (Nielsen & Ganter, 2018), the effects of commercial technology platforms on social life (Bucher, 2018; van Dijck, 2013), and the ways that Facebook and Google blend the features of commercial platforms and public infrastructures (Plantin, Lagoze, Edwards, & Sandvig, 2018). Specifically in the domain of politics, our previous study (Kreiss & McGregor, 2018) revealed how technology firms shape political communication through the provision of customer service around their platforms, especially in the context of political advertising, as firms are motivated to work in the political space for the purposes of revenue, marketing, and access to the political field. Related are works such as Vaidhyanathan's (2018) critical look at how Facebook routinely shapes the political sphere through its commercial logics, which incentivizes things that have been at the forefront of scholarly concern since the 2016 U.S. presidential election: "fake news," targeted advertising, clickbait, and emotional outrage.
In the field of political communication, there has been an explosion of work on digital advertising. Fowler (2018) and West (2017) provide sweeping overviews of the contemporary landscape of political advertising, including on television and social media, detailing the key regulatory frameworks, economic contexts, and practices behind it. There are a number of journalistic reports that detail aspects of the digital advertising ecosystem, including changes since the 2016 U.S. presidential election in terms of verification and transparency (see Wells & Seetharaman, 2017) and the advertising targeting that these platforms afford (for a summary see Kaye, 2017). There has been increasing attention to the capacity of Facebook to afford psychological and personality-based targeting given the Cambridge Analytica scandal (Sharp, Danenberg, & Bellman, 2018) that in turn has revealed much about how Facebook's "custom audiences" and advertising algorithms function and shape exposure (see Eckles, Gordon, & Johnson, 2018), although Anstead, Magalhaes, Stupart, and Tambini (2018) challenge the dominant narratives of advertising segmentation. Kim et al.'s (2018) important research provides an analysis of how during the 2016 U.S. presidential election groups not registered with the FEC, including Russian interests, ran divisive issue campaigns in swing states, facilitated by the agency's own failure to create effective rules for digital political advertising. Edelson, Sakhuja, Dey, and McCoy (2018) use Facebook's new political advertising database implemented after the 2016 U.S. presidential election to detail aggregate patterns of digital political advertising spend and the political actors behind it.
Despite this growing body of literature, there is little scholarly work on the political advertising policies and practices that Facebook and Google have, which ultimately determine what paid political messages voters actually see. Given this, we specifically seek to build on Gillespie's (2018) sweeping look at platform content moderation and analyze these practices in the context of political advertising. Gillespie's book is the most comprehensive analysis to-date in the literature on platforms as "custodians" of public speech through the work of content moderation that they perform. Gillespie focuses on categories of speech that all platforms generally regulate in some way (sexual and graphic
4
Daniel Kreiss and Shannon C. Mcgregor
content, harassment, hate speech, illegal activity, and self-harm) and shows how platforms are challenged by the problem of "scale" which results in three approaches: editorial review (oversight of content through employees at these firms), community moderators (users themselves flag problematic content), and detection through computational methods.
While Gillespie (2018, p. 203) does not analyze content moderation in the context of advertising specifically or discuss politics as a special domain of practice, generally he argues that on most platforms paid advertising content has more specific, explicit, and stringent rules than regular content. To our knowledge, no research has explicitly analyzed the rules around paid political content in the U.S., or their origins, evolution, and enforcement. This paper therefore fills an important gap in the literature in offering an analysis of the policies and processes that Google and Facebook have with respect to paid political content, with an eye towards how advertising policies are defined, interpreted in technical and human practice, and enforced. As such, this paper asks: What processes, guided by what policies, govern the paid political content on Facebook and Google? What rationales do Facebook and Google provide for action - and inaction - in regulating paid political content on their sites? What ability do campaigns and other organizations have to contest the decisions of these firms in regulating their content?
Methods
To answer these questions, we conducted interviews with seven former employees of Facebook (n = 2) and Google (n = 4) that were involved in or knowledgable about elections-related work (See Appendix A for a list of interviewees, general descriptions of their roles, and interview dates).1 We identified these individuals through snowball sampling - contacts that we already had in the political field based on our previous work recommended them. Given ongoing professional relationships with these firms, these interviews are presented not-for-attribution. These interviews lasted approximately an hour each and followed a semi-structured interview guide organized around questions related to their work at these firms and the political advertising approval process and enforcement. Of the interviewees, four were explicitly familiar with our previous work (Kreiss & McGregor, 2018) and that became the express starting point for discussion.
To gain the perspective of the political practitioners who have to routinely work with Facebook and Google, and provide a check on our interview findings with these former firm employees, we conducted interviews with 11 political practitioners from campaigns, political organizations, and digital political consultancies. These interviews included seven former campaign staffers from the 2016 presidential cycle and five national digital political consultants who work routinely with these companies. Taken together they have worked in digital for electoral campaigns at all levels of office during the 2012, 2014, and 2016 cycles and in roles ranging from senior directors overseeing digital operations to political advertising specific positions. For the presidential staffers, the first author inserted questions specific to this study into a separate project that analyzed the experiences of women (n = 45) working in political technology (See Kreiss & Adams, 2019). In addition, we conducted a largely off-the-record site visit to a prominent national political consultancy to hear and see first-hand about the approval process and rejected ads, which served to confirm many of the findings here. There was no standard interview guide for
Arbiters of What Voters See
5
this study. Building off of our previous work (Kreiss & McGregor, 2018), all interviews were specifically tailored to the individual after extensive preparation, including reviewing Linkedln career history, public writings, and press appearances, and work on specific campaigns or for organizations. That said, we did approach each with some common broad themes that focused on the ad approval process, specific cases when content was rejected, and appeals made to these companies, as well as adapted questions iteratively as we learned more to test findings and hypotheses (see Small, 2009). Because of the need for political practitioners to maintain good working relationships with Facebook and Google, and an often-expressed wariness of being negative about these firms, all interviews were conducted not-for-attribution. In addition, we circulated a draft paper to five of our interviewees who were particularly insightful and invited comments and critiques -we received extensive comments from three individuals and conducted a third interview with a fourth to clarify policy-related matters.
We also analyze emails (10 threads, 45 emails total) exchanged between Facebook government and elections staffers and staffers of a U.S. 2017 gubernatorial and 2016 presidential campaign. We provide analysis of the emails here with the permission of one of the people in these exchanges, who we do not identify and who provided them to the first author of this study. In each case of the 10 separate threads, the email exchanges contain the full discussion of the specific issue the campaign staffers raised and its resolution (or lack thereof) of as it played out over email. As such, we do not have additional data - such as phone calls or conversations between Facebook and other campaign staff - where these issues may have been discussed. To check the validity of our interpretation of the data here, we sent the draft of this paper for comments to the individual who provided these emails and conducted a review of journalistic accounts of these races to confirm details. Because of the sensitive nature of this data, the first author redacted the identifying dates, locations, and people referred to in the emails and noted them in text. In addition, in the interest of having a fuller range of perspectives and evidence to bear on the findings presented here, as well as provide the other parties to these emails a chance to respond, we circulated an earlier draft of this paper to a senior-level individual working in this domain at Facebook and asked for comments. We used subsequent exchanges with this individual and other materials to contextualize our findings.
The UNC Institutional Review Board approved this study. The IRB's terms provide explicit permission to report materials collected both in interviews and in the course of field observation, including at political events, "and the offices of social media firms and campaign organizations during the 2016 presidential election and 2018 midterm cycle."
Data from these interviews and email exchanges provide direct evidence on how Facebook and Google regulated paid political content through 2017 and early 2018. We note where we believe changes have occurred since then, especially as Facebook and Google have sought to respond to congressional and public pressure.2 Regardless of the fact that this article offers a snapshot of a particular moment in time, as we conclude in the discussion section the broader tensions over transparency, public justification, and negotiations over content remain, even as specific cases continue to arise and platform companies evolve in response. At the same time, we also acknowledge that both Facebook and Google have rolled out a number of new initiatives, and have expanded existing ones, since the time of writing this piece that we cannot address in depth here and will have to be considered in subsequent work.3
6
Daniel Kreiss and Shannon C. Mcgregor
Findings: The Policies and Processes behind Political Advertising at Google and Facebook
In this section, we detail some of the explicit policies that guide political advertising on Google and the processes through which they are enforced and implemented, while drawing comparisons with Facebook.4 Google is an ideal test case because political practitioners perceive it to have the clearest policies around paid political advertisements and paid speech among the platform companies. While the literature often groups firms such as Facebook and Google together under the broad label of "platforms," in the eyes of political users they are not equivalent. In interviews, political users routinely distinguished between Facebook and Google, particularly in the context of the latter's comparatively more developed policies, processes, and organizational structure for dealing with content-related decisions. As one former Google employee and current political practitioner stated, Google was: "Not sophisticated... [but] further along of all of them.... in terms of having teams, and the build-out, and the development. They've just always, because they're older, you can kind of always tell they're five years older" (Former Google Staffer A). A Republican practitioner echoed this (Republican Digital Consultant A), stating that Facebook seemed to be a presidential cycle behind Google in developing its content policies.
Google has both general advertising rules and industry-specific rules. General rules relate to the format and style of ads. In addition, advertising has to comply with national laws (which look different across different industries) and the company supports targeting based on location, age, gender, and parental status (demographic categories), affinity (interests or characteristics), consumer behavior or search, specific sites, similar (or lookalike) audiences, and remarketing to users who have already interacted with an ad, in addition to content targeting.5 Like Google, Facebook has general advertising policies regulating format in addition to those that apply explicitly to paid political content. Facebook offers many more categories into which users can be sorted and targeted, including: location (state, zip code, or congressional district), demographics, age, gender, languages spoken, relationship status, education level, work status and place of employment, income, "ethnic affinity," generation, life events, politics, Facebook connections, plus a wide array of other interests and tracked behaviors online.6 In addition, at the time of this writing Facebook permitted targeting on the basis of likelihood to engage with political content and based on ideology (a scale from very conservative to very liberal). Both Facebook and Google enable campaigns and other advertisers to leverage their own data for the purposes of targeting through "custom audiences" (Facebook) and "customer match" (Google).
At the same time, both Facebook and Google provide broad, general guidelines relating to all permissible advertising when it comes to content, which detail matters of "policy and principle" and reveal "how platforms see themselves as ambivalent arbiters of public propriety" (Gillespie, 2018, pp. 46^47). For example, Google provides broad definitions of what types of advertising content can run on its platform, and that apply to politics.7 In addition to "prohibitions against counterfeit goods," "dangerous products or services," and "enabling dishonest behavior," there is a broad category of "inappropriate content":
Inappropriate content
We value diversity and respect for others, and we strive to avoid offending users, so we don't allow ads or destinations that display shocking content or promote hatred, intolerance, discrimination, or violence.
Arbiters of What Voters See
7
Examples of inappropriate or offensive content: bullying or intimidation of an individual or group, racial discrimination, hate group paraphernalia, graphic crime scene or accident images, cruelty to animals, murder, self-harm, extortion or blackmail, sale or trade of endangered species, ads using profane language.8
Exactly what constitutes "intimidation" or "discrimination" is not specified, however, other than what Google itself will determine these things are in specific cases (which in turn will recursively feed back into the guidelines, which means that they change over time). Similarly, Facebook provides the general guideline that all content appearing on its platform, including paid political ads, must meet "community standards." These standards are meant to be comprehensive, rooted in the following principles: safety, voice, and equity. Facebook asserts that their policies around community standards are based on input from "experts in fields such as technology and public safety." While Facebook did publish a 27-page document providing the rationale for many of their standards in April 2018, it is not transparent about the process of their creation, including what experts were consulted.9
In their very flexibility, these broad guidelines provide a framework for these firms to work within as they approve or disapprove paid advertisements. For example, at Google, the Ad Policy team establishes the policies that govern advertising. The first level of vetting advertising content is an algorithmic filter that screens for certain types of content. The content that is flagged by this algorithm then goes to a human pool of reviewers. As a former Google staffer (Former Google Staffer B) with campaign experience related:
All ads go through algorithmic filter.... Algorithmic filter says out of 100 ads, 20 of these look kind of weird. 10 of them are definitely weird, flag, shut down. 10 of them look a little bit weird. Okay, up. Human giant pool of reviewers who are applying policy from their own judgment that they have received from even one layer up, which are the people in Mountain View who make the rules. You've got advertiser, artificial layer, human pool, people who make the rules.
Most political advertising at Google does not go before a human. Those advertisements that do are flagged by the initial algorithmic process. When content gets flagged for review, a team of dedicated employees apply ad policies to specific, flagged ads to decide what should be approved and not. Former Google staffers argued that this team tries to apply these policies as consistently and fairly as possible. Former staffers, in turn, cite that specific cases of ads feed back into the principles that govern all political advertising. As one former Google staffer who worked with campaigns (Former Google Staffer C) described:
Then if there's recurring problems that occur over the course of, say a year, then we might go to the people that set the policy and be like, "Hey, we'd love to have a conversation about why this is the case and show you some instances where we think this might be inconsistently applied and then have a conversation about it."
8
Daniel Kreiss and Shannon C. Mcgregor
This is important because the content guidelines of a firm such as Google are sweeping. Some make seemingly arbitrary distinctions. For example, one former Google staffer (Former Google Staffer C) pointed to an ad that was disapproved because it called a candidate a "liar" - "which was not allowed but, you could say that 'he lied' or something like that. There's weird nuances like that..." Others relate to substantive matters of policy that significantly limit the capacity of political actors to contest public issues. Another former Google and campaign staffer described how ads relating to the gun control debate might run afoul of Google's prohibition against "inappropriate content." Speaking from the perspective of being a campaign staffer, this former Google employee (Former Google Staffer D) relates how:
We got our ads disapproved all the time, all the time.... That was hard. That was hard for us because we would go back and say, "Look, these are topics that are worthy of public discourse. They're worthy of having a debate about, and you're saying we can't say this because it's too provocative..." Let's say we wanted to have a debate about AR-15s right now, in the wake of all these mass shootings. There's a really good chance that a display ad that got placed by an anti-gun group or a pro-gun group would get disapproved. Then you'd have to go back and have those conversations. "Well, you can't advertise about guns because it's violence and violence is a protected category." You say, "Well, we should actually have laxer standards about this because we should be having a public conversation about guns and this is one way in which we can do that." There was always a back and forth about what should go and what should not go.
Not specific to Google, political practitioners pointed to having similar issues running advertisements on platforms relating to issues such as immigration and abortion. The challenge is that these firms have the powerful ability to shape paid messages in the public sphere, but rely on vague and flexible policies using terms such as "protected category" or "inflammatory" and "inappropriate" content.10
At the same time, determining where to draw the line is deeply fraught and an institutionally subjective process - where the outcomes are shaped by a broad set of commercial and legal imperatives, managerial decisions, and the capacity for internal deliberation within firms such as Google and Facebook, informed by arguments with external actors. Former Google and Facebook staffers point to how complicated questions are when it comes to political advertising. As one former Google staffer (Former Google Staffer D) who subsequently went on to work on political campaigns detailed:
People think that, oh, Google and Facebook, they're so thoughtless. They let these things slip by and then they wait until it's crazy. They wait until something blows up in their face and then they change it. That's not true. Again.. .at least on Google's end, people are really trying to do the right thing and think deeply about weighing the balance between 'we want to be for free expression and allow both parties to have a public debate about important topics' with 'we want to protect people from things that are really offensive or shocking and violent or things like that.' I've been in those conversations. People are very, very thoughtful about it. It's hard, right? That's the other thing. Whenever you see simplistic answers to these questions, that person's
Arbiters of What Voters See
9
not being serious. That's just not a serious person because they're really hard circles to square. You can quote me on that. It's not a serious person who thinks that these are easy problems to solve.
This points to the internal deliberation that occurs within these companies. Former Google staffers, for instance, cited internal debates that occurred on a regular basis about not only paid content, but the use of sites such as YouTube for political speech more broadly. The first author witnessed this first-hand at a platform company in the context of a debate over non-paid content, where employees articulated different perspectives on content some perceived as harassing in the context of gun policy. Our interviews made clear that thinking of these firms as one unified actor, as a monolithic platform, is misleading; they are composed of individuals and teams with different perspectives and views that come to bear during debates over paid and other content. As this former Google staffer relates (Former Google Staffer D), controversial decisions often come down to an internal deliberation, or at times more accurately, debate where different perspectives clash.
For example, while most political ads are self-serve, meaning that any registered user can purchase them directly on Google, other campaigns or political entities are "managed service accounts" and have dedicated staffers who can advocate for them in-house. Both Facebook and Google have dedicated staffers who work with campaigns along partisan lines (Kreiss & McGregor, 2018.) These staffers play an advisory role regarding what advertising may get approved and what may not in advance of an ad being submitted or run. Former campaign staffers cited that managed service teams at Google can help facilitate or expedite the ad approval process, but they do not make ad approval decisions themselves.11 As one former Google staffer (Former Google Staffer C) described how the process worked during and after the 2012 election cycle:
Typically approvals at Google used to take between 24 to 48 hours, sometimes 72 which is too long, especially in politics. So frequently we would get, you know hundreds of emails a week with just a customer ID from our clients, clients that we know, that are in our book of business, and they would say 'could you approve ads in this account?' And they'd send the customer ID. That customer ID then we'd have to file, or we used to have to... I don't know how they do it now but we used to have to file a ticket internally with that customer ID under expedited rules and they would get approved or disapproved within four hours by an ad approvals team....
And so the approvals became a headache sometimes when like, I remember certain ads that like were a sleeve, where someone was wearing cutoffs, you know a cutoff shirt where their arms were showing and that would get disapproved for like having too much skin. And you have to go back and war with like the ad approver on why it should be approved or not.
While these staffers do not make approval decisions directly, they can internally advocate on behalf of their clients. Echoing this staffer, former Google staffers stated that members of sales teams can make a case for their clients with their ad policy colleagues. As such, across these interviews, it was clear that part of the negotiation process over approved ads also happens internally at Google. As a former Google staffer (Former Google Staffer B) described, echoing the staffer above:
10
Daniel Kreiss and Shannon C. Mcgregor
Well, we do have, I mean there were guidelines and so whenever something got rejected they would flag the specific policy. Right like, oh 'this was disapproved for this policy, here's a link to it.' We would read the policy, we may interpret it differently and we would have an argument, or we would try to escalate it up if we adamantly disagreed with the disapproval of that ad.
Former Google staffers were at pains to argue that any "escalating" done on behalf of a client came in the context of existing policies and guidelines. In essence, that this is not about bending the rules for a client, or creating exceptions to these rules, but about disagreements over the interpretation of particular ads or policies. As a former Google staffer (Former Google Staffer C) described it, the process is about a good faith effort to have objectivity and fairness in the implementation of existing rules and guidelines. As a different former Google staffer (Former Google Staffer A) echoed: "A typical conversation might be: 'You disapproved this ad because X, but actually the rule says X is ok' NOT 'you disapproved this ad because X, and X is not permitted, but X SHOULD be permitted..
At Facebook, the stated flow for paid political content is that all ads are reviewed, and usually approved, within 24 h. Official steps for what to do when an ad is disapproved include editing the ad (although it is clear from our conversations with informants that Facebook does not always provide enough feedback to understand why the ad was disapproved and what might be changed to comply). In addition, ads can be appealed through a form on the Facebook ad platform.12 These guidelines were implemented in May of 2018.13 In addition, political professionals at a large agency we visited for this research in summer of 2018 also described to us a process wherein they can appeal ad disapprovals to the Facebook account managers they work with, who can then escalate the issue within the company. These political professionals themselves said this process might not be fair to smaller companies or campaigns without account managers, although the process remained for them a "black box" (i.e.: they do not know what happens when appeals are escalated, such as which teams it then goes to, what the processes for adjudication are, and whether it is more likely to result in approval). As such, the work flows and decision-making around approvals and disapprovals of political content is not at all clear, even to practitioners themselves.
For political practitioners, the internal processes at Google were also not transparent. One digital consultant (Republican Digital Consultant A) cited, and furnished examples of, "elevating" rejected advertisements to their sales representatives at Google. After that, the advertisements were approved, but with nothing in the way of an explanation as to what shifted in the policy or its interpretation by employees. At the same time, higher-level campaigns and digital political consultants with multiple clients also cited arguing their cases directly around ad disapprovals. On campaigns at the scale of a presidential and within digital agencies working with multiple campaigns, practitioners stated that they have voice vis-a-vis these firms and can at times influence their decision-making, even in the context of a broad lack of transparency over the approval process. As this consultant (Republican Digital Consultant A) described his interactions with platform companies:
I think the real bad ones are the ones that are one-layer...like the Verizons, and AOL. We had a campaign the other day for a client. We gave them [dozens] of pieces of ad content and they rejected all [of them], because they talked about illegal immigration, putting up a wall.... This concept of the
Arbiters of What Voters See
11
'inflammatory,' they're really giving young ad approvals people, like, the determination of what goes up and what doesn't.
And I say that because Google and those guys tend to do it a little bit better, but sometimes you catch some. But, like, all the ad tech stuff, this is becoming a huge problem of they're not letting you run creatives that they deem inflammatory, but that's all they said: "It's inflammatory." Doesn't say, like, why?
So, we spend a lot of our time, on behalf of our clients, actually getting on the phone, and getting up to three levels of above the person who says, "No." Cause my whole argument is, like, "We're having this fight now because in August we're not going to ask you again, we're just going to take it straight to the press and point this out."
As is clear from this quote, digital consultancies and campaign staffers can and do make arguments about why advertisements should be approved to Google employees. Given the interpretative flexibility of Google's guidelines, and as importantly the many gray areas in politics, the potential for a deliberative process around approvals exists, but as currently constituted it lacks transparency and Google has the final say without much in the way of public accountability save campaigns going to the press. As a former staffer of state-level campaigns (Democratic Digital Political Staffer C) described a debate that she had with Google over an ad, in this case whether it was factually true:
The Google process does take longer [than Facebook]... .We had an ad that we were trying to run for a candidate that was about the transgender ban, and they said that it wasn't technically a ban. We got into a long argument essentially with their policy team about the language and whether that was an accurate characterization of it. That was completely different than my experience with Facebook. Never dealt with any issues with Facebook, but Google actually was more active in terms of vetting the content... .1 think it was the campaigns or the issues team, but it definitely was a real person, and I talked with the person on the phone and went through a bunch of stats to get the content processed.
The fact that this staffer could not recall exactly the person or team she was dealing with regarding approvals points to a larger issue brought up across our interviews: the lack of clear processes for adjudicating paid content disputes, fueled at least in part by the lack of clear organizational charts detailing jurisdiction and employee responsibilities for either Google or Facebook. As one political practitioner (Republican Digital Political Staffer B) stated in the context of Facebook: "At one point, I think it was earlier this year, I said can I see an org chart?... I was dying to get my hands on one because at one point, I was trying to work on something and I was like I don't know who's really in charge here. Which, I think, is personally in my personal opinion is that's a little bit of a frustration point with working with them."
At both of these firms, the lack of clear justification for content decisions and processes for appeals is particularly concerning given that, in our view, voice is important for improving firm editorial decision-making. Without this mechanism, Google and Facebook would have the authority to determine the messages that campaigns pay for the public to see, without the challenges that keep these firms learning and cause their
12
Daniel Kreiss and Shannon C. Mcgregor
policies to evolve. In the next section, we turn to the complex ways that speech was negotiated in practice at Facebook during an election and show the clear limits of the voice that practitioners have onto the decision-making of these firms.
Negotiations Over Speech in Practice at Facebook
To illustrate how negotiations over paid speech play out over the course of a campaign, here we examine email exchanges between campaign staffers and firm employees. These email exchanges reveal a) how technical and content decisions are often intertwined, b) how Facebook struggles to reconcile its public statements with its internal policies, c) the active debate that at times occurs between Facebook and campaign staffers over paid political speech on the platform, and d) the lack of clear public disclosure or transparency around Facebook's active regulation of the affordances of its platform and paid speech. We focus here on an exchange between a Republican 2017 gubernatorial campaign staffer and a Facebook staffer over a rival campaign's use of edited newspaper headlines in paid promotions. Facebook would go on to ban the practice in September 2017, but at the time of this exchange, advertisers on the platform could edit the headlines and descriptions of publishers or other generators of content on the platform (Cohen, 2017). While this exchange provides a window onto a case during the course of one election, it illustrates these four general points.
The exchange between the campaign and Facebook staffers started when a campaign staffer argued that a rival changed the headline of an article from The Washington Post in a misrepresenting way in a paid promotion and asked what remedy the campaign had. In response a Facebook staffer argued that:
re: the ability to edit headlines/descriptions, thank you for flagging; this is something we're working to address and I'll share this example with the appropriate team.
Re: fighting back - harkening back to my old job, if it's creating blowback then this is where I'd personally start:
1) consider commenting on a post like this as [the candidate] - something simple. "Not true.... [redacted]" Something that's more in-line and on-message for him, but if it's causing trouble and this is a community you want to engage, confront it. If not from [the candidate], then maybe someone else on the campaign - but good if it's from his verified presence. Now, does that invite angry people there to come and engage on your Page (and potentially leave negative comments)? 100%, yes. But if they do that, that gives you the chance to correct them there as well.
2) also - if you have something on your Page or website of [the candidate]/ the campaign talking about the issue, link to that in the comments. Or provide to supporters to share themselves.
3) maybe the WaPo reporter who wrote the article, or Politifact [redacted], would write up a post about the claim (especially if your opponent has repeated it somewhere)? These obviously elevate the issue; not sure that's what you want or not.
Arbiters of What Voters See
13
4) add a positive, evergreen version of the issue to your "issues" tab once you launch that - not going to make an impact right now, but will be there to point to (and have others point to).
5) start up a separate Page/website you use to debunk and set the record straight when needed. HRC had a Page called, "The Briefing" last year: https://www.facebook.com/TheBriefing2016/rve seen others do similar things. Might be a valuable platform to have now and in a general election. I obviously defer to you on what topics you feel the need to engage on or ignore, but if I was going to confront an issue this is where I'd start.
I hope this is helpful.
There are a couple of salient points here in Facebook's response to this complaint. First, that Facebook's guidance on how to deal with the issue of misrepresentation essentially comes down to "more speech." This is a deeply agonistic view of political debate; the Facebook employee does not want to accept any editorial role for the firm. Instead, it is incumbent on the rival campaign, the target of misinformation, to challenge it through "fighting back" - having the candidate rebut the charge or a third party respond to it, such as a fact checker with a legacy press organization.
In response, the gubernatorial staffer points out the degree to which Facebook's advice is ultimately self-serving in that, left unsaid by the firm's representative, it requires paid advertising to effectively engage in. At the same time, the campaign staffer expresses frustration about the fact that the rival campaign is violating Facebook's own policies, the need to address issues like this takes time and energy that could be better spent, and the company has a near monopoly on reaching a majority of their candidate's supporters:
Glad to hear y'all are taking steps to fix this problem. I think it will be a great step. Low hanging fruit in my opinion. Any ideas on when it will be implemented? It's helpful to have for links you own (for ads, etc.), but in this case it's caused a firestorm for our campaign that could have been avoided. As always, I'm happy to provide whatever feedback or insights that I can to your teams to help improve the product for candidates. I don't feel like I've ever been taken up on this offer.
All good points on fighting back which I've been working on most of this afternoon (instead of more productive items, like raising money to pay for more Facebook ads promoting great videos or engaging petitions). I've posted from [candidate's] page into the comments for that doctored post, but it's buried and there's nothing to stop them from deleting the comment.
The page says it's a 501 c4 registered organization, but we've been unable to find any evidence to that effect. Does that affect anything on your end that they're misrepresenting their tax status?
You will recall we did have the [redacted] page in [redacted] that did OK and we may be doing it again, but as you know that takes a lot of advertising to do effectively - promoting the page, building a following, promoting content, etc. and just seems pointless if pages are going to be allowed to make doctored posts like this. We're having to pay more money in Facebook ads to combat fake news on Facebook. It seems like a racket.
14
Daniel Kreiss and Shannon C. Mcgregor
I don't like the idea of having to counter with our own fake news but it might be the only option.
Again, I want to reiterate that this has been an ongoing headache because an anonymous page posted a Washington Post article and gave it a fake headline. This is as cut and dry a case of fake news that I've ever seen.
As always, this isn't personal and I respect you and consider you a friend, I just don't feel like legitimate criticisms are being heard or acted upon. And we have no other recourse because 76% of our target voters are on the platform.
This exchange continued over the course of a month. It escalated as the rival campaign paid to promote the edited Washington Post headline. Ultimately, Facebook made the decision to remove the posts with the edited headline without any stated rationale, seeming change in policy, or accountability for the decision - either to the campaign that raised the complaint or the broader public. About a month after the initial complaint, a Facebook staffer simply emailed the campaign to say: "I wanted to circle back and let you know we were able to remove the post with the edited headline. We saw they re-posted it with the same edited headline and have already flagged that as well. If you spot anything else like this going forward, please send our way and we'll look into it asap." Checking news reports at the time, it appears a reporter's question about the altered headline may have prompted the take down; a spokesman cited that it violated Facebook's prohibition in its Terms of Service against doing: "anything unlawful, misleading, malicious, or discriminatory."14
A senior staffer at Facebook (Facebook Staffer A) stated that internally this particular case was more complicated than this explanation. The staffer stated that Facebook was internally actively debating the technical ability to edit headlines at the time this issue was flagged by the campaign, and made the decision in the middle of the 2017 election cycle to change the functionality of the platform. The ability to edit headlines in paid promotions was designed to enable news organizations to test their headlines and maximize engagement. What was clear, however, is that this affordance could be abused by other, strategically motivated actors. Because there was an active internal debate over a potential policy change at the firm, Facebook staffers did not discuss it with any of their campaign clients (and, if they did, they would have to discuss it with all of them, making it de facto public). In essence, the staffer stated that the decision was not a question of a content violation at all, but a universal change in policy relating to the technical affordances of the platform that happened to coincide with this election.15
Again, however, according to the documents we reviewed, there was not a rationale or attempt at one at the time for the campaign staffer who initially flagged the issue. Meanwhile, the senior Facebook staffer argued that the spokesperson was referring to the fact that the editing of headlines is "misleading" on its face because it changes what publishers originally wrote - and as such this was not a judgment on the truth of the content. While this is technically true, and was the same as other public statements about the shift in the platform's policy at this time (Campbell, 2017), the lack of additional context behind this sudden policy change ultimately provided little insight into the internal impetus and logic behind a universal technical change that the company claimed was not a content-based decision. Ultimately, we believe that the campaign staffer and the public were left wondering why this practice was suddenly in violation of Facebook's terms. Broadly, how or why the decision was made was anyone's guess at that moment in time, outside of Facebook.
Arbiters of What Voters See
15
The issue surfaced again less than a month later when the rival campaign began using the same edited headlines, previously taken down by Facebook, in their videos on the platform. This time, the Facebook staffer working with the campaign provided a response to the campaign that was consistent with the technical explanation cited above. When asked by the campaign staffer "if you can explain to me the reason why Facebook would remove the link with the fake headline, but not a video or screenshot?" the Facebook employee responded that:
The difference here is that our policy around editing of headlines on link shares is based on misuse of a product feature, not on determining whether an edit is true or false. People cannot use our product functionality to edit an article headline unless they are the original publisher of that content. This is an objective stance, and as you can see in our Page terms it does not take into account whether an edit is misleading or deceptive in nature - it's simply not allowed.16
As I've mentioned before, we are not arbiters of truth - we empower the community and third-party fact checkers to determine that for themselves. I know I've outlined these and other strategies for you before, but if you believe a piece of content is not true, you should consider working with third-party fact checkers to get them to review - or, if you believe something is defamatory and want to pursue legal action against it, I strongly encourage you to report it to our Intellectual Property team at ip@fb.com.
Again, we'll always have reviewed any specific item you send to us. We're here to do all we can to help, even if the result in a particular case is that we can't take any action.
The Facebook staffer's response reveals a backing away from the language of "misleading" and justification of the policy in terms of applying to a "misuse of a product feature," which is "objective" in the company's eyes. On questions of "truth," the firm's political staffer states that on procedural grounds third parties have to do that work. Indeed, in a later email the Facebook staffer states that it is "reports from the Facebook community that dictate when third party fact checking organizations make a determination about whether something is true or false," which ultimately could lead to a "disputed" label, a new feature the firm implemented in December 2016 in response to fake news and that the campaign staffer desired.17 That said, in response to the Facebook staffer's email about not being an "arbiter of truth", the campaign staffer points to the structural conditions of the media ecosystem that Facebook should consider before abnegating its responsibility over what appears on its platform. Specifically, that Facebook fails to consider the limited recourse to fact-checking or publicity that many state and congressional candidates have in a world of large declines in local news (in part facilitated by Facebook's growing share of digital advertising). Even more, the staffer points out Facebook's power as an "arbiter of attention", and the fact that the firm is acting less responsibly than television and radio outlets, with more extensive paid content approval processes. Finally, the staffer argues that the recommendations from Facebook about how to combat these ads in essence result in a "shakedown" for more advertising dollars:
16
Daniel Kreiss and Shannon C. Mcgregor
As I've mentioned before, there are no fact checkers in this race. Politifact [regional] shut down last year. There are only three statewide reporters covering this campaign and their combined readership is less than the reach of [candidate's] Facebook posts.
While you may argue that Facebook is not the arbiter of truth, you are the arbiter of what our voters see. Numerous third parties that we have been able to get to write about this have said it's untrue and egregiously so. Radio stations have taken down the ad and TV stations would too if [redacted] tries to run it there. Why won't Facebook?
Facebook has a moral responsibility to ensure that voters receive accurate information.
We simply refuse to participate in the shakedown of having to pay for our own ads to counter this nonsense. Our response to the lie will never be as incendiary as the initial lie itself.
Echoing other campaign staffers, one frustration was that while these arguments were taking place behind-the-scenes, Facebook founder and CEO Mark Zuckerberg was publicly discussing, lamenting, and promising to address the fact that the platform was a hotbed of misinformation. During the 2017 gubernatorial campaign, for instance, numerous third-party journalistic outlets had debunked claims of this campaign's opponent, yet this staffer seemingly could not get Facebook's alert applied to this case (or the content to be removed). In response, a Facebook staffer cited that fact-checked warnings have to originate from the Facebook "community" (without specifying what that entails) and, most confusingly, that the "disputed" content feature "is for reporting links to stories on third-party sites, not on-platform videos or other posts, so it wouldn't apply in this case." As the Facebook staffer goes on to relate:
I want to be clear again that we're not arbiters of truth. We want to give people a voice; we won't decide for ourselves whether a political candidate's statements are accurate or inaccurate.
As on any topic, if you have articles making your case from fact checkers and others, I encourage you to share them - from assets you control, but also get supporters to post them/comment with them too. You have a strong grassroots army - activate them if you think it'd be valuable. Counter-speech is an important (and large) part of what happens on the platform.
In our estimation, this exchange reveals an incoherent set of standards at play regarding Facebook's policies with respect to speech. On one level, Facebook is saying that it is not going to be an "arbiter of truth" and therefore is going to focus on "engaging both our community and third party organizations."18 And yet, when the press is involved, a content take-down is publicly justified on substantive grounds (i.e.: a violation of terms of service prohibiting "anything unlawful, misleading, malicious, or discriminatory.") This, in turn, appears to be at odds with the actual reason for the takedown, a universal policy decision about a technical affordance, not content. This fact was used to justify to the campaign staffer why subsequent inaction for the very same content in a different form is justified on procedural grounds. Regardless, all the while, Mark Zuckerberg and other Facebook representatives were speaking publicly with a clear epistemological and editorial stance in the context of banning "misleading" content,
Arbiters of What Voters See
17
"hoaxes," and "fake news," even as there is a complicated set of procedures that removes themselves of the responsibility for determining what content falls in these categories and therefore allowing it to remain on the platform. In response to this last email from Facebook, the campaign staffer cited Zuckerberg's own words: '"We don't want any hoaxes on Facebook. Our goal is to show people the content they will find most meaningful, and people want accurate news.' - Mark Zuckerberg."
Discussion: The Tension between Arbiters of Attention and Truth
As the email exchanges between campaign and Facebook staffers presented above make clear, there is a fundamental tension at play between Facebook's desire to be an arbiter of political attention and avoid being an arbiter of political truth. In terms of the former, Facebook has significant commercial incentives to be an arbiter of attention both in terms of immediate revenues, such as direct political advertising sales, and increased user engagement which in turn leads to greater revenue over time. Facebook also has considerable field incentives to remain the key arbiter of political attention in terms of its long-term growth and position in a competitive marketplace. Finally, Facebook has considerable regulatory incentives to remain an arbiter of attention given the degree to which the company has explicitly used its rapid growth strategy as a way to forestall regulatory efforts (see Vaidhyanathan, 2018).
At the same time, Facebook has clear disincentives to be an arbiter of truth, even as increasing public pressure calls on the company to fulfill this role. The problem of "truth" is manifold. First, as Facebook's staffers rightfully pointed out, the question of what is a political truth is, as Lucas Graves (2016) has illustrated, a deeply contested and challenging one. The firm will likely not be able to adjudicate political truth at scale and certainly not at the speed that has made the platform so attractive for political users. Second, the company clearly lacks both the expertise and knowledge to perform this democratic function, and likely the legitimacy to do so. Indeed, the company will be hard-pressed to put itself in the position of adjudicating the claims of elected representatives and those vying for public office who, after all, are carrying their appeals directly to the public. More generally, Facebook is downstream from politics, and as such is challenged to respond to what is a broader crisis in public discourse and contests of civic epistemology (see Kreiss, 2017).
The tension between Facebook's role as an arbiter of political attention and its relationship to political truth, to our eyes, cannot be resolved. At best, there are possible ways to balance competing concerns and interests. To take one example from the exchanges above: Currently, Facebook's business model is entangled with its policies around paid political speech. The advice to the campaign staffer to combat what in our estimation was a clear case of political misinformation was more paid political speech and greater use of the platform. Both of these things benefit Facebook's bottom line. One solution here is to potentially remove the affordances that make it easy for political actors to manipulate and doctor texts that did not originate with them, which in this case Facebook did (but not with an eye to all the possible permutations and without a clear rationale and public justification). Beyond this, Facebook could subsidize counter-speech in cases where the company, or its partners in things such as fact checking, have determined that mis- or dis-information is at play. Facebook could not only make its judgments
18
Daniel Kreiss and Shannon C. Mcgregor
clear and public, but also provide additional means for entities affected by this speech to counter it through non-revenue means, such as free advertising or an allowed rebuttal appended to the offending content. This will help take the financial and human resources burden off of the targets of mis- and dis-information. At the same time, the disincentivization of speech the company deems harmful is already happening if we believe Facebook's own claims to using machine learning to limit the reach and engagement of things such as deliberately fake news - itself a clear example of the way that Facebook is an arbiter of political discourse. One problem is that the criteria Facebook has for doing so is not transparent and the company's actions, which entail questions of interpretive meaning hidden beneath the legitimating language of machine learning, collapse a political problem into a technical and engineering problem. Even further, machine learning decisions are often made without any public rationale or justification.
In this formulation, in the context of the exchanges above, Facebook can take a middle ground: the company only has to identify which speech it sees as potentially problematic and be clear and public as to the grounds upon which it is making that determination, not judge the epistemological status of the claims itself. The company only has to empower other actors to a right of rebuttal or reply, not determine the content of that reply. In other words, the company will make a determination between which speech it respects and which is potentially problematic, not necessarily weigh in on the truth claims of political speech.
More generally, however, all of this is contingent on these platform companies developing more public, transparent, clearer, and better justified policies and processes around their editorial roles with respect to paid political speech. We do not believe that there is a universal content-based framework for making speech decisions outside of egregious content, the takedown of which is not seriously disputed (Gillespie, 2018). In a world without moderation, the most vile human speech, from child pornography to explicit racist imagery, would reign free. The questions, as always, are how the lines are drawn and what constitutes permissible content. The easiest case in some ways is paid political advertising - no one seriously disputes that these firms have the right to regulate purchased speech on their platforms. As such, beyond the forms of subsidy we detail above, we end here with outlining more transparent and contestable policies and procedures by which platform editorial decisions can be made and, ultimately, contested by users exercising voice. While Facebook, for instance, has provided greater clarity around its community standards and has implemented greater transparency in its political ads, such as compelling disclosure among advertisers and for advertising, we focus here on policies around advertising and the appeals process.19 In our view, these things will strengthen the decision-making of these firms and ultimately further democratic speech.
In political scientist Albert Hirschman's (1970, p. 43) "exit, voice, and loyalty" framework, his concern was with "the invention of such institutions and mechanisms as can communicate complaints cheaply and effectively" - what he saw as the "art of voice." Our stance is that to cultivate voice, the decisions and processes that Facebook and Google have regarding paid content moderation need to be transparent to political users themselves; there needs to be clearer and more public justifications for the editorial decisions the firms make as well as institutionalized channels to dispute specific decisions and policies; and, finally, a wider public voice onto the workings of these firms should be
Arbiters of What Voters See
19
encouraged given the public's role as a stakeholder - and by "public" we mean the many different groups in pluralist society.
As our study reveals, one significant problem in the context of paid content moderation is that editorial decision-making often occurs without disclosure or justification to the practitioners affected or the public at large. When decisions are made in the course of a campaign, rationales are not clear and can be conflicting depending on who is citing them. There is also no clear institutional means to challenge decisions. This is a significant failure of the firms that maintain private platforms for public speech. At a minimum, when campaigns or digital consultancies submit paid advertising for approval, if it runs afoul of a platform firm's advertising policies for reasons such as "inappropriate content" or a violation of "community standards," there should be clear justifications provided to campaigns for that rejection based on specific content as well as formal channels for contesting these decisions.
We view as inevitable too that there will be different levels of appeal afforded to and disclosure provided to political actors, Facebook users, and the public at large. At best, these firms should seek to create a more transparent and ultimately contestable process for all of their paid political users, under the theory that adversarial processes between campaigns (such as rivals), parties, and these firms will help provide for accountability within the system. While this may be a goal, we recognize the inherent challenges of letting stakeholders, such as campaigns, and by default the public, know about internal policy debates during the course of elections. But what is clear from the email exchanges above is that Facebook has more to do when it comes to explaining the processes it has, and decisions it does make, to its campaign stakeholders and the public. And, there was no clear process for a stakeholder to contest a decision like the one documented above. In an ideal world, outside actors such as journalists can be empowered to provide additional layers of scrutiny over the decisions that these firms make (such as when to reject, limit the reach of, or subsidize speech). At the same time, the issue of scale simply prevents the broader community of Facebook users or the public at large from substantially weighing in on these issues.
At a basic level, then, these firms should provide more detailed rationales regarding their editorial decision-making and be more transparent about how it works in practice, including publicly justifying the decisions they make. This has clearly not been the case. Most of what we know about Facebook's content moderation (in the context of both paid and "organic" speech), for instance, has come from leaked documents obtained by journalists.20 Facebook and Google have a duty to explain the way that their content moderation works and to make clear and defend their decisions to their broader user communities and the public at large, and revise them in response to public pressure with clear rationales and justifications. We suspect that public pressure from their user communities and the public at large will only come in relation to larger issues of significant public attention, and that fits with their "monitorial role" in democratic life (Schudson, 1999).
At the same time, we understand that revenues are at play and providing differential levels of access to these firms is likely routine practice in commercial advertising. But, in democratic processes, these practices raise clear concerns about equity and fairness for political advertisers, who after all are vying for political office or contesting public issues. For example, it is clear that in the current way political advertising is conducted, some political actors have more access to Facebook and Google and therefore greater opportunities to contest speech than other actors or the public. As the cases of both Facebook
20
Daniel Kreiss and Shannon C. Mcgregor
and Google reveal, the political consultancies handling multiple accounts, as well as presidential campaigns, bring greater revenue to these companies and have more ready access to account representatives that can facilitate their appeals processes - but not intervene in editorial decisions directly.
While the Facebook email exchanges above are but one example, staffers also cite that they routinely interact with campaigns on a number of different levels, from explaining policies and offering potential remedies to at times making policy changes based on both internal feedback and external public pressure. These processes can be more formalized and transparent to the actors themselves, and to this end, we believe the proposed Facebook content oversight board given more specific public delineation in January 2019 is a welcome step.21 In the political context, in the interest of fairness, campaigns that work outside of the consultancy system or that are too small to have significant digital advertising budgets should have the same ability to contest decisions as other, larger actors.
Finally, beyond transparency and the ability to contest the decisions that firms make, there needs to be stronger deliberative processes internal to these firms. As this paper revealed, these firms are not homogenous entities. They are composed of staffers with differing views and perspectives on political speech working in various parts of their sprawling organizations. And, they debate courses of action with respect to content moderation. These firms are also not hermetically sealed entities; they are responsive to public pressure and scrutiny. Given these things, Facebook and Google should purpo-sively create the conditions for robust internal debates, and ones that are informed by the exercise of voice outside of their walls. Perhaps they do already - it is impossible to know. But it does not appear as such.
Conclusion
From school board seats to The White House, platforms developed and maintained by firms such as Facebook and Google provide important means by which officeholders and those seeking office speak to and engage with the electorate. This paper shows how while Facebook and Google resist being labeled as arbiters of political speech, they actively vet the paid content that appears on their platforms, with significant implications for the resulting political discourse on their platforms. These decisions often occur in a black box - there is little justification or explanation for campaigns, leaving these firms open to perceptions of political bias. Facebook and Google should institutionalize more robust forms of political voice in the future.
Acknowledgement
The authors wish to thank Bridget Barrett for her comments on an earlier draft of the study.
Disclosure statement
No potential conflict of interest was reported by the authors.
Arbiters of What Voters See
21
Notes
1. For Google, we focus here on Google's ad network and advertising policies, not YouTube or its other properties. The Google Ads Policies can be found here: https://support.google.com/ adspolicy/answer/6008942?visit_id=636856675945721223-1772300979&rd=l (accessed 21 13/19).
2. Since the 2016 U.S. presidential election, both Facebook and Google implemented verification processes to ensure the authenticity of those buying political advertisements. For Google, to purchase Google Ads around politics accounts need to have verification through the FEC or a tax ID (see https://support.google.com/adspolicy/answer/9002729?hl=en&ref to-pic=1316596 (accessed 2/13/19). Facebook requires verification through a postal address to run ads related to a list of political topics that Facebook provides (see: https://www.facebook. com/business/help/208949576550051 (accessed 2/13/19)).
3. This includes Facebook's political ads archive, launched in June 2018 and a new political ads verification process, launched in April 2018.
4. For Facebook, these advertising policies apply to Instagram, Facebook Messenger, and the company's broader audience network.
5. Google, Targeting Your Ads. Available online at: https://support.google.com/google-ads /answer/1704368?hl=en (accessed 2/11/19).
6. Although investigations from ProPublica have found that these tools can be used to target users whose "interests" include white nationalism or exclude users from minority racial identities. https://www.propublica.org/article/facebook-advertising-discrimination-housing-race-sex-national-origin.
7. Google Ads Policies. Available online at: https://support.google.com/adspolicy/answer/ 6008942?visit id=l-636682172203706782-1677536111&rd=l (accessed 2/13/19).
8. Google, Inappropriate Content. Available online at: https://support.google.com/adspolicy/ answer/6015406?hl=en (accessed 10/12/18).
9. https://www.facebook.com/communitystandards (accessed 2/12/19). For a review of the update of its community standards and appeals process in April 2018 see: https://www.npr. org/2018/04/24/605107093/facebook-updates-community-standards-expands-appeals-process.
10. In the wake of revelations that Russian actors created fake pages and accounts to post political content about the 2016 U.S. presidential election, Facebook created an authorization process that applies to anyone wanting to post ads related to candidates running for office, political parties, PACs, or advocates for specific electoral outcomes; related to election, referendum, or ballot initiatives (including get out the vote ads); is regulated as political advertising; or relates to any "national legislative issue of public importance." In the U.S., these issues are broad and far-ranging, including: abortion, budget, crime, economy, guns, energy, infrastructure, health, taxes, and values. We have heard from our informants - and there are accounts in the press of this as well - of Facebook's issue-related ad policy being applied to all sorts of non-political ads. Ads relating to these issues must, per Facebook's terms, include a paid-for-by disclaimer that appears to users targeted with the post. However, Facebook does not verify that these paid-for-by disclaimers are accurate. After the requirement, news organizations have shown that nearly anything can be put into the self-regulated "paid-for-by" input, and these ads, with false disclaimers, were approved to run on the platform. Facebook recently announced that they are rolling out these stricter rules in countries with upcoming elections, including India, Nigeria, Ukraine, and the EU. At the time of this writing, the loophole in disclaimer requirements in the U.S. was not closed, and it is unclear as to what the impact of the varying rules across countries has had on attempts to undermine disclosure requirements.
11. The Google elections team, which is made up of account managers and executives organized under "sales," do not make decisions about what ads should be approved and what should not.
22
Daniel Kreiss and Shannon C. Mcgregor
Former staffers were at pains to draw distinctions between the elections team, as a sales team, and the ad policy team.
12. https://www.facebook.com/business/help/313752069181919.
13. https://newsroom.fb.com/news/2018/05/ads-with-political-content/.
14. Facebook Terms of Service. Available online at: https://www.facebook.com/terms.php.
15. Facebook puts this in the category of regulating content not based on what an actor is saying, but in essence how they are saying it. This is similar to the category of how Facebook regulates pages based on behavior (such as during the midterms, taking down pages that evidenced spamming behavior or that used fake accounts).
16. It appears this policy change received public attention only months after this email. It cannot be independently verified when the policy change actually occurred.
17. Facebook, Addressing Hoaxes and Fake News. Available online at: https://newsroom.fb.com/ news/2016/12/news-feed-fyi-addressing-hoaxes-and-fake-news/.
18. Facebook, Addressing Hoaxes and Fake News. Available online at: https://newsroom.fb.com/ news/2016/12/news-feed-fyi-addressing-hoaxes-and-fake-news/ Going further, in 2018 Facebook did start taking action on content instead of just behavior (such as behaving in similar ways to spam profiles), to no longer permitting accounts that produced something fact checkers said was false to run advertising.
19. Facebook updated its community standards and expanded its appeals process in April 2018, see: https://www.npr.org/2018/04/24/605107093/facebook-updates-community-standards-expands-appeals-process.
20. See "The Impossible Job: Inside Facebook's Struggle to Moderate Two Billion People." Motherboard. 2018, August 23. Available online at:https://motherboard.vice.com/en_us/arti-cle/xwk9zd/how-facebook-content-moderation-works.
21. See "Charting a Course for an Oversight Board for Content Decisions," available online at: https://newsroom.fb.com/news/2019/01/oversight-board/ (accessed 2/12/29).
References
Anstead, N., Magalhäes, J. C, Stupart, R., & Tambini, D. (2018). Politcal advertising on Facebook: The case of the 2017 United Kingdom General Election. Presented to the American Political Science Association, Annual Meeting, Boston.
Bucher, T. (2018). If... then: Algorithmic power and politics. New York, NY: Oxford University Press.
Campbell, C. (2017, March 9). NC senate leader phil berger changes news headlines on facebook posts - Violating the site's policy. News & Observer. Retrieved from https://www.news-observer.com/news/politics-government/state-politics/article 13 7442113 .html
Cohen, D. (2017, September 14). Advertisers on Facebook can no longer edit news articles' headlines, descriptions. AdWeek. Retrieved from https://www.adweek.com/digital/advertisers-on-facebook-can-no-longer-edit-news-articles-headlines-descriptions/
Eckles, D., Gordon, B. R., & Johnson, G. A. (2018). Field studies of psychologically targeted ads face threats to internal validity. Proceedings of the National Academy of Sciences, 115(23), E5254-E5255.
Edelson, L., Sakhuja, S., Dey, R., & McCoy, D. (2018). An analysis of United States online political advertising. arXiv.org. Retrieved from https://arxiv.org/abs/1902.04385
Fowler, E. F. (2018). Political advertising in the United States. Philadelphia, PA: Routledge.
Gillespie, T. (2018). Custodians of the Internet: Platforms, content moderation, and the hidden decisions that shape social media. New Haven, CT: Yale University Press.
Graves, L. (2016). Deciding what's true: The rise ofpolitical fact-checking in American journalism. New York, NY: Columbia University Press.
Arbiters of What Voters See
23
Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations, and
states (Vol. 25). Cambridge, MA: Harvard University Press. Kaye, K. (2017, January 3). Data-driven targeting creates huge 2016 political ad shift. AdAge.
Retrieved from: https://adage.com/article/media/2016-political-broadcast-tv-spend-20-cable-
52/307346
Kim, Y. M., Hsu, J., Neiman, D., Kou, C, Bankston, L., Kim, S. Y, ... Raskutti, G. (2018). The stealth media? Groups and targets behind divisive issue campaigns on Facebook. Political Communication, 35(4), 1-27.
Kreiss, D. (2017). The fragmenting of the civil sphere: How partisan identity shapes the moral
evaluation of candidates and epistemology. American Journal of Cultural Sociology, 5(3),
443-459. doi:10.1057/s41290-017-0039-5 Kreiss, D., & Adams, K. (2019). Navigating the brogrammers and the boys' club: Women's
representation and experiences in political technology. New Media & Society, doi: 10.1177/
1461444819835573
Kreiss, D., & McGregor, S. C. (2018). Technology firms shape political communication: The work
of Microsoft, Facebook, Twitter, and Google with campaigns during the 2016 US presidential
cycle. Political Communication, 35(2), 155-177. doi:10.1080/10584609.2017.1364814 Newell, J., & Blevins, J. L. (2018). Transparency in political advertising: Assessing the utility and
validity of the FCC's Online public inspection file system. Journal of Information Policy, 8,
417. doi:10.5325/jinfopoli.8.2018.0417 Nielsen, R. K., & Ganter, S. A. (2018). Dealing with digital intermediaries: A case study of the
relations between publishers and platforms. New Media & Society, 20(4), 1600-1617.
doi:10.1177/1461444817701318 Plantin, J. C, Lagoze, C, Edwards, P. N, & Sandvig, C. (2018). Infrastructure studies meet platform
studies in the age of Google and Facebook. New Media & Society, 20(1), 293-310. doi: 10.1177/
1461444816661553
Schudson, M. (1999). The good citizen: A history of American civic life. Cambridge, MA: Harvard University Press.
Sharp, B., Danenberg, N., & Bellman, S. (2018). Psychological targeting. Proceedings of the National Academy of Sciences, 115(34), E7890-E7890. doi:10.1073/pnas.l810436115
Small, M. L. (2009). How many cases do I need? On science and the logic of case selection in field-based research. Ethnography, 10(1), 5-38. doi: 10.1177/1466138108099586
Soper, S. (2018, September 19). Amazon increases ad market share at expense of Google, Facebook. Bloomberg. Retrieved from https://www.bloomberg.com/news/articles/2018-09-19/amazon-increases-ad-market-share-at-expense-of-google-facebook
Vaidhyanathan, S. (2018). Antisocial media: How Facebook disconnects us and undermines democracy. New York, NY: Oxford University Press.
van Dijck, J. (2013). The culture of connectivity: A critical history of social media. New York, USA: Oxford University Press.
Wells, G., & Seetharaman, D. (2017, October 13). Facebook users were unwitting targets of Russia-backed scheme. The Wall Street Journal. Retrieved from https://www.wsj.com/articles/face-book-users-were-unwitting-targets-of-russia-backed-scheme-1507918659
West, D. M. (2017). Air wars: Television advertising and social media in election campaigns, 1952-2016. Washington DC: CQ Press.
24 Daniel Kreiss and Shannon C. Mcgregor
Appendix A.  Interview Table
Role Identification Role Description Date(s) of Interview
Former Facebook Staffer A	Analytics and monetization	10/23/17 and 1/5
Former Facebook Staffer B	Advertising	9/29/17
Facebook Staffer A	Senior Management	2/8/19
Former Google Staffer A	Politics, Google and Campaign Consulting	11/3/2017
Former Google Staffer B	Politics, Google and Campaign Staff	1/26/18
Former Google Staffer C	Politics, Google	11/8/17
Former Google Staffer D	Technology, Google and Campaign Staff	3/26/18
Republican Digital Consultant A	Technology Consulting and Campaign Staff, Presidential	2/15/18; 8/15/18
	(2016)	
Republican Digital Consultant B	Technology Consulting and Campaign Staff, Presidential	3/6/18
	(2016) and State Races	
Republican Digital Consultant C	Technology Consulting and Campaign Staff, Presidential	3/26/18
	(2016)	
Republican Digital Political Staffer A	Political Organization Staff	1/23/17
Repbulican Digital Political Staffer B	Technology Consulting and Campaign Staff, Presidential	11/14/17
	(2016)	
Republican Digital Political Staffer C	Technology Consulting and Political Organization Staff	1/23/17; 3/26/18
Democratic Digital Political Staffer A	Campaign Staff, Presidential (2016)	11/8/17
Democratic Digital Political Staffer B	Campaign Staff, Presidential (2016)	3/6/18; 2/21/17
Democratic Digital Political Staffer C	Campaign Staff, Gubernatorial, House Races	11/8/17
Democratic Digital Political Staffer D	Campaign Staff, Presidential (2016)	12/12/17
Democratic Digital Political	Campaign Staff (numerous presidential bids and state level	11/27/17
Consultant E	campaigns)