Division as Revenue
Abstract
Two businesses in American politics are paid for attention and urgency. Social platforms sell advertising against engagement, and fundraising conduits and campaigns collect gifts through emails, recurring-gift defaults and deadlines. This chapter takes each in turn and covers both parties wherever the record allows. Meta's filings state that substantially all of its revenue is advertising and that falling engagement is a risk to it, and its 2017 annual report said its 2018 News Feed change would reduce time spent. Observational studies find that out-group posts and moral-emotional language spread more, and a 2025 field experiment on X found that reranking hostile posts moved out-party feelings. Randomized experiments run with Meta in 2020 found no significant change in polarization from changing feeds, removing reshares, reducing like-minded sources or deactivating accounts. Twitter's own audit and two YouTube studies add findings that cut against a simple story. In fundraising, WinRed refunded 7.4 percent of its 2020 individual contributions and ActBlue 2.3 percent, a Republican-site study estimated that pre-checked recurring boxes added over $43 million, and an email study found manipulative tactics across parties. The Democratic-side record on defaults is thinner and partly reported. Whether engagement ranking causes lasting polarization is unresolved. No cited document records an intent by a platform executive or a campaign.
3.1 What Meta’s filings state
Meta's annual reports state that substantially all of its revenue comes from selling advertising, and its 2017 report lists decreased user engagement and time spent as a risk to that revenue. Advertising revenue was $39.9 billion in 2017, $55.0 billion in 2018 and $160.6 billion in 2024.
Evidence: Exhibit 712, Exhibit 713, Exhibit 714
| Fiscal year | Advertising revenue |
|---|---|
| 2017 | $39,942 million |
| 2018 | $55,013 million |
| 2024 | $160,633 million |
The reports used are Meta's Form 10-Ks for fiscal years 2017, 2018 and 2024. Advertising was 98 percent of the company's $40,653 million in revenue in 2017, according to the 2017 report's risk factors. The 2024 report gives advertising of $131,948 million for 2023 and $113,642 million for 2022.
The filings describe how Meta earns money. They give no revenue by type of content, and no filing compares revenue from divisive posts with revenue from other posts. Press reports quote a post of 5 October 2021 in which Mark Zuckerberg said that advertisers consistently tell the company they do not want their ads next to harmful or angry content.
Exhibit 712 · SEC filing
Facebook, Inc. Form 10-K for fiscal year 2017
we recently announced changes to our News Feed ranking algorithm to help our users have more meaningful social interactions, and we expect that these changes will have the effect of reducing time spent and some measures of user engagement with Facebook, which could adversely affect our financial results.
Annual report. Advertising $39,942M of $40,653M revenue (98% of revenue per risk factors). Lists decreased user engagement, including time spent, as a risk to ad revenue, and says the announced News Feed change was expected to reduce time spent.
- Source
- Facebook, Inc. (SEC filing)
- Date
- 1 February 2018
- Quoted from
- Item 1A, Risk Factors (user engagement risk factor)
- Grade
- Primary record
- Document
- https://www.sec.gov/Archives/edgar/data/1326801/000132680118000009/fb-12312017x10k.htm
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 713 · SEC filing
Facebook, Inc. Form 10-K for fiscal year 2018
We generate substantially all of our revenue from selling advertising placements to marketers.
Annual report. Advertising $55,013M of $55,838M revenue in 2018. Repeats that News Feed changes for more meaningful interactions have had the effect of reducing time spent.
- Source
- Facebook, Inc. (SEC filing)
- Date
- 31 January 2019
- Quoted from
- Item 1, Business (Overview)
- Grade
- Primary record
- Document
- https://www.sec.gov/Archives/edgar/data/1326801/000132680119000009/fb-12312018x10k.htm
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 714 · SEC filing
Meta Platforms, Inc. Form 10-K for fiscal year 2024
Currently, we generate substantially all of our revenue from selling advertising placements on our family of apps to marketers
Annual report. Advertising $160,633M of $164,501M revenue in 2024 ($131,948M in 2023; $113,642M in 2022).
- Source
- Meta Platforms, Inc. (SEC filing)
- Date
- 30 January 2025
- Quoted from
- Item 1, Business (Overview)
- Grade
- Primary record
- Document
- https://www.sec.gov/Archives/edgar/data/1326801/000132680125000017/meta-20241231.htm
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.2 The 2018 News Feed change
In January 2018 Adam Mosseri announced that Facebook would prioritize posts that spark conversations and meaningful interactions between people in News Feed. The company's 10-K for 2017 said it expected the change to reduce time spent and some measures of user engagement.
Evidence: Exhibit 715, Exhibit 712, Exhibit 713
Mosseri's post is dated 11 January 2018 and warns that Pages may see reach and watch time decrease. The 10-K was filed on 1 February 2018. The 10-K for 2018 repeats that the changes had the effect of reducing time spent.
Facebook's stated reason, in a May 2020 post by its vice president of integrity, was research showing that people find more meaning in content from people they know than in passively consuming content. Reports of internal documents, covered in section 3.4, say the heavier weighting of comments and reshares in this ranking rewarded contentious posts. Facebook disputes that account.
Exhibit 715 · Company announcement
Bringing People Closer Together
With this update, we will also prioritize posts that spark conversations and meaningful interactions between people.
Announcement of the News Feed ranking change that prioritizes posts that spark conversations and meaningful interactions; warns Pages may see reach and watch time decrease.
- Source
- Facebook (Adam Mosseri, Head of News Feed)
- Date
- 11 January 2018
- Quoted from
- Body, second section
- Grade
- Primary record
- Document
- https://about.fb.com/news/2018/01/news-feed-fyi-bringing-people-closer-together/
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.3 The Haugen statement and the replies
Frances Haugen, a former Facebook product manager, told a Senate subcommittee in a written statement of 4 October 2021 that the company's choices produced "a system that amplifies division, extremism, and polarization" and that its internal research confirmed the problems. Facebook's vice president of integrity had said in May 2020 that the company took steps against polarization, some at a cost to revenue, and Zuckerberg said on 5 October 2021 that the profit-over-safety claim was untrue.
Evidence: Exhibit 716, Exhibit 717, Exhibit 718
The official position
Facebook's vice president of integrity answered the Wall Street Journal's May 2020 report that the company had shelved polarization research. He wrote that the story "wilfully ignored critical facts that undermined its narrative" and listed the 2018 News Feed change, integrity staffing, recommendation restrictions and hate-speech enforcement. Zuckerberg called the profit-over-safety claim untrue.
What the record shows
Haugen's statement characterizes Facebook leaders' choices and says internal research confirmed the problems. The statement does not reproduce the research. Each side's account is a document of its own position, and the cited records contain no regulator or court finding on the dispute.
The statement is Haugen's account of the documents. It is dated 4 October 2021, and the hearing followed on 5 October. Press reports quote Zuckerberg's reply of 5 October. The company's published replies and Haugen's statement are each a party's own account, and neither is a finding of fact.
Exhibit 716 · Congressional testimony (written statement)
Statement of Frances Haugen, Senate Commerce Subcommittee on Consumer Protection, Product Safety, and Data Security
The result has been a system that amplifies division, extremism, and polarization — and undermining societies around the world.
Former Facebook product manager's written statement. Characterizes Facebook leadership's choices and states that internal research confirmed the problems. A statement of her account; the underlying documents are not reproduced in it.
- Source
- Frances Haugen (U.S. Senate Committee on Commerce, Science, and Transportation)
- Date
- 4 October 2021
- Quoted from
- p. 2
- Grade
- Primary record
- Document
- https://www.commerce.senate.gov/index.php/services/files/FC8A558E-824E-4914-BEDB-3A7B1190BD49
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 717 · Company statement
Investments to Fight Polarization
this particular story wilfully ignored critical facts that undermined its narrative.
Facebook's response the day after the Wall Street Journal reported it shelved polarization research. Lists the 2018 News Feed change, integrity staffing, recommendation restrictions and hate-speech enforcement.
- Source
- Facebook (Guy Rosen, VP Integrity)
- Date
- 27 May 2020
- Quoted from
- Opening paragraph
- Grade
- Primary record
- Document
- https://about.fb.com/news/2020/05/investments-to-fight-polarization/
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 718 · Executive statement (reported)
Mark Zuckerberg's public response to the Haugen testimony, as reported
Zuckerberg's post responding to the Senate hearing. States that advertisers tell the company they do not want their ads next to harmful or angry content, and that the claim that Facebook puts profit over safety is untrue.
- Source
- Mark Zuckerberg, Facebook (reported by SiliconANGLE)
- Date
- 5 October 2021
- Grade
- Reported
- Document
- https://siliconangle.com/2021/10/05/facebooks-mark-zuckerberg-denies-company-focuses-growth-safety/
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.4 Documents reported by the press
News reports of internal Facebook documents state that staff warned in 2018 and 2019 that ranking changes rewarded outrage and that company leaders declined some proposed changes. Facebook disputes these accounts.
Evidence: Exhibit 716, Exhibit 717, Exhibit 718
The reports come from the Wall Street Journal and a consortium of news organizations, and concern an internal analysis of whether the platform rewards outrage, a 2018 staff presentation on divisiveness and a 2019 staff account of complaints from European political parties about the incentives created by the 2018 change.
The documents reach the public through Haugen, who has said she copied tens of thousands of pages. Congress received redacted copies, and the Securities and Exchange Commission received her filings. The published Facebook responses cited here do not deny that the documents exist. The cited records contain no SEC finding on them.
Staff research and discussion are not company decisions, and a staff warning does not show what leadership concluded.
3.5 What spreads
In 2,730,215 Facebook and Twitter posts from news media and members of Congress, posts about the political out-group were shared or retweeted about twice as often as posts about the in-group. Twitter studies find that positive social feedback increases later expressions of moral outrage and that moral-emotional language increases diffusion within ideological groups.
Evidence: Exhibit 719, Exhibit 720, Exhibit 721
Rathje, Van Bavel and van der Linden published the first result in PNAS on 23 June 2021. Their summary states that out-group language was the strongest predictor of shares, and the authors state that their results are correlational. Out-group language also predicts shares on news-media accounts, so the pattern is not specific to one platform.
Brady and colleagues published a Twitter study of gun control, same-sex marriage and climate change messages in PNAS in 2017. Their 2021 Science Advances paper reports two preregistered observational studies covering 7,331 users and 12.7 million tweets, and two behavioral experiments with 240 participants. The authors state that the observational studies cannot support causal inference.
These studies describe what spreads on the platforms. Their authors do not claim that the ranking systems caused the patterns, and the sample does not separate the effect of ranking from the effect of what people choose to share.
Exhibit 719 · Peer-reviewed study
Out-group animosity drives engagement on social media
we found that posts about the political out-group were shared or retweeted about twice as often as posts about the in-group.
Analysis of 2,730,215 Facebook and Twitter posts from news media and US congressional members. Out-group language was the strongest predictor of shares; authors state results are correlational.
- Source
- Rathje, Van Bavel, van der Linden (PNAS)
- Date
- 23 June 2021
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.2024292118
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 720 · Peer-reviewed study
Emotion shapes the diffusion of moralized content in social networks
the presence of moral-emotional language in political messages substantially increases their diffusion within (and less so between) ideological group boundaries.
Twitter study on gun control, same-sex marriage and climate change messages.
- Source
- Brady, Wills, Jost, Tucker, Van Bavel (PNAS)
- Date
- 26 June 2017
- Quoted from
- Significance statement
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.1618923114
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 721 · Peer-reviewed study
How social learning amplifies moral outrage expression in online social networks
positive social feedback for outrage expressions increases the likelihood of future outrage expressions, consistent with principles of reinforcement learning.
Two preregistered observational Twitter studies (7,331 users, 12.7 million tweets) and two behavioral experiments (N = 240). Authors state the observational studies cannot support causal inference.
- Source
- Brady, McLoughlin, Doan, Crockett (Science Advances)
- Date
- 13 August 2021
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/sciadv.abe5641
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.6 Two audits of amplification
Twitter's own audit found that the mainstream political right was amplified more than the mainstream left in six of seven countries and found no evidence that algorithms amplify extreme groups more than moderate ones. An analysis of 208 million U.S. Facebook users found that ideological segregation rises from potential exposure to engagement, and that most misinformation identified by Meta's fact-checking program sat in a homogeneously conservative corner with no liberal equivalent.
Evidence: Exhibit 722, Exhibit 727
Huszar and colleagues, Twitter's own researchers, published their audit in PNAS (online 21 December 2021). They compared about 2 million accounts on a reverse-chronological feed with accounts on the ranked feed. Germany was the seventh country, where the difference was not significant. Right-leaning U.S. news sources were amplified marginally more. The authors call the causal mechanism a topic for further study and write that amplification does not imply promotion of extreme ideology.
The Facebook analysis, by Gonzalez-Bailon and colleagues in Science on 28 July 2023, belongs to the same group of 2020 election studies done with Meta as section 3.8. It uses aggregated 2020 data. The two audits differ in what they measure, and each finding holds for the platform and period it covers.
Exhibit 722 · Peer-reviewed study
Algorithmic amplification of politics on Twitter
In six out of seven countries studied, the mainstream political right enjoys higher algorithmic amplification than the mainstream political left.
Audit using a holdout group of about 2 million accounts on a reverse-chronological feed. Right amplified more than left in six of seven countries (Germany not significant); right-leaning US news amplified marginally more; no evidence extreme groups were amplified more than moderate ones. Authors call the causal mechanism a topic for further study.
- Source
- Huszar, Ktena, O'Brien, Belli, Schlaikjer, Hardt (PNAS; Twitter authors)
- Date
- 21 December 2021
- Quoted from
- Abstract (PMC8740571)
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.2025334119
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 727 · Peer-reviewed study
Asymmetric ideological segregation in exposure to political news on Facebook
most misinformation, as identified by Meta’s Third-Party Fact-Checking Program, exists within this homogeneously conservative corner, which has no equivalent on the liberal side.
Aggregated data for 208 million US Facebook users, 2020. Segregation rises from potential exposure to actual exposure to engagement; conservative-only news corner holds most fact-checked misinformation.
- Source
- Gonzalez-Bailon, Lazer, Barbera, Zhang, Allcott, Brown et al. (Science)
- Date
- 28 July 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/science.ade7138
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.7 Twitter’s engagement ranking
A preregistered audit of Twitter's engagement-based ranking, published in PNAS Nexus in 2025, found that it amplified emotionally charged, out-group hostile content that users said made them feel worse about their political out-group.
Evidence: Exhibit 723
Milli and colleagues compared the engagement-based ranking with a reverse-chronological baseline and surveyed users. They report trade-offs in an alternative that ranks by stated preference, including possible reinforcement of content users already agree with.
The result concerns what the ranking promotes and how users say they feel. It does not measure a lasting change in attitudes.
Exhibit 723 · Peer-reviewed study
Engagement, user satisfaction, and the amplification of divisive content on social media
Twitter’s engagement-based ranking algorithm amplifies emotionally charged, out-group hostile content that users say makes them feel worse about their political out-group.
Preregistered audit of Twitter's engagement-based ranking against a reverse-chronological baseline, with user surveys.
- Source
- Milli, Carroll, Wang, Pandey, Zhao, Dragan (PNAS Nexus)
- Date
- 27 February 2025
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1093/pnasnexus/pgaf062
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.8 The 2020 experiments with Meta
In randomized experiments run with Meta during the 2020 election, switching users to chronological feeds, removing reshares, reducing exposure to like-minded sources and deactivating accounts changed what users saw and produced no significant change in polarization.
Evidence: Exhibit 724, Exhibit 725, Exhibit 726, Exhibit 728, Exhibit 740
Guess and colleagues moved consenting Facebook and Instagram users to reverse-chronological feeds for three months. The chronological feed did not significantly alter issue polarization, affective polarization, political knowledge or other key attitudes. A second Guess study removed reshares from Facebook feeds and found no significant effect on polarization or any individual-level political attitude measure.
Nyhan and colleagues cut like-minded-source exposure by about one-third for 23,377 users and found no measurable effect on eight preregistered attitude measures. Allcott and colleagues deactivated Facebook for 19,857 users and Instagram for 15,585 users for six weeks before the election. Their estimates for affective and issue polarization, perceived legitimacy of the election, candidate favorability and turnout were close to zero.
The studies were randomized, preregistered and large. Their windows ran from six weeks to three months, they cover one election, and they measure attitudes. They do not measure effects on the people who produce content or on political norms over years. The authors state that Meta cooperated on data and design and that the company did not have prepublication approval.
A September 2024 Science eLetter by Bagchi and colleagues argued that the control condition of the feed experiment overlapped temporary Meta emergency measures, which changed what the comparison group saw. The Guess authors replied that the control condition had changed and that the core result stood, and Science did not ask for corrections. The eLetter is not peer reviewed, and university summaries report it.
Exhibit 724 · Peer-reviewed study
How do social media feed algorithms affect attitudes and behavior in an election campaign?
the chronological feed did not significantly alter levels of issue polarization, affective polarization, political knowledge, or other key attitudes during the 3-month study period.
Randomized experiment with Meta cooperation: consenting Facebook and Instagram users moved to reverse-chronological feeds, 3 months, 2020 election. Study window overlapped temporary Meta emergency measures (criticized in a 2024 Science eLetter).
- Source
- Guess, Malhotra, Pan, Barbera, Allcott, Brown et al. (Science)
- Date
- 28 July 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/science.abp9364
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 725 · Peer-reviewed study
Reshares on social media amplify political news but do not detectably affect beliefs or opinions
Contrary to expectations, the treatment does not significantly affect political polarization or any measure of individual-level political attitudes.
Randomized removal of reshares from Facebook feeds during the 2020 election.
- Source
- Guess, Malhotra, Pan, Barbera, Allcott et al. (Science)
- Date
- 28 July 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/science.add8424
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 726 · Peer-reviewed study
Like-minded sources on Facebook are prevalent but not polarizing
had no measurable effects on eight preregistered attitudinal measures such as affective polarization, ideological extremity, candidate evaluations and belief in false claims.
Field experiment, 23,377 users; like-minded-source exposure cut by about one-third.
- Source
- Nyhan, Settle, Thorson, Wojcieszak, Barbera, Chen et al. (Nature)
- Date
- 27 July 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1038/s41586-023-06297-w
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 728 · Peer-reviewed study
The effects of Facebook and Instagram on the 2020 election: A deactivation experiment
the effects of both Facebook and Instagram deactivation on affective and issue polarization, perceived legitimacy of the election, candidate favorability, and voter turnout were all precisely estimated and close to zero.
Randomized 6-week deactivation of 19,857 Facebook and 15,585 Instagram users before the 2020 election.
- Source
- Allcott, Gentzkow, Mason, Wilkins, Barbera et al. (PNAS)
- Date
- 13 May 2024
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.2321584121
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 740 · Published critique (eLetter, not peer reviewed)
Social media algorithms can curb misinformation, but do they? (Science eLetter)
Argues the Guess et al. control condition overlapped temporary Meta emergency measures. Guess et al. replied; Science did not request corrections.
- Source
- Bagchi, Menczer, Grabowicz et al. (Science eLetters)
- Date
- September 2024
- Grade
- Reported
- Document
- https://osome.iu.edu/research/blog/science-eletter-social-media-algorithms-can-curb-misinformation-but-do-they
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.9 The 2024 experiment on X
A preregistered 10-day field experiment on X in 2024 with 1,256 participants, using a browser-side reranking method independent of the platform, found that reranking posts with partisan animosity shifted out-party feelings by more than 2 points on a 100-point scale.
Evidence: Exhibit 729
Piccardi and colleagues published the study in Science on 27 November 2025. The authors describe it as providing causal evidence that exposure to partisan-animosity content alters affective polarization. The shift is a little over 2 points on a 100-point scale after 10 days.
The design differs from the 2023 Meta experiments. Piccardi changed which posts appeared by their content, and the Meta experiments changed how the feed was structured. The result is consistent in direction with the observational findings in section 3.5.
Exhibit 729 · Peer-reviewed study
Reranking partisan animosity in algorithmic social media feeds alters affective polarization
providing causal evidence that exposure to AAPA content alters affective polarization.
Preregistered 10-day field experiment, 1,256 participants on X during the 2024 campaign, with an independent browser-side reranking method.
- Source
- Piccardi, Saveski, Jia, Hancock et al. (Science)
- Date
- 27 November 2025
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/science.adu5584
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.10 YouTube
Two peer-reviewed studies of YouTube found little evidence that its recommendation algorithm drove attention to radical content, and found that exposure to alternative and extremist channels came through subscriptions and external links among users who already held high levels of resentment.
Evidence: Exhibit 730, Exhibit 731
Hosseinmardi and colleagues used user-trajectory data across the political spectrum for 2016 to 2019 and published in PNAS on 2 August 2021. Chen and colleagues combined behavioral and survey data for 1,181 people in 2020 and published in Science Advances in September 2023. They write that YouTube's algorithms were not sending people down rabbit holes during their 2020 observation window.
The Chen authors note that YouTube changed its recommender system in 2019 and that their 2020 result may reflect those changes.
Exhibit 730 · Peer-reviewed study
Examining the consumption of radical content on YouTube
We find little evidence that the YouTube recommendation algorithm is driving attention to this content.
User-trajectory data across the political spectrum, 2016-2019.
- Source
- Hosseinmardi, Ghasemian, Clauset, Mobius, Rothschild, Watts (PNAS)
- Date
- 2 August 2021
- Quoted from
- Significance statement
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.2101967118
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 731 · Peer-reviewed study
Subscriptions and external links help drive resentful users to alternative and extremist YouTube channels
YouTube’s algorithms were not sending people down “rabbit holes” during our observation window in 2020
Paired behavioral and survey data, n = 1,181, 2020.
- Source
- Chen, Nyhan, Reifler, Robertson, Wilson (Science Advances)
- Date
- September 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1126/sciadv.add8080
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.11 What remains unresolved
Whether engagement-optimizing ranking causes lasting polarization, as distinct from what it amplifies and what users see, is unresolved in the peer-reviewed record.
Evidence: Exhibit 724, Exhibit 729, Exhibit 723
The 2023 experiments found no detectable attitude change over six weeks to three months. The 2024 X experiment found a small change over 10 days. Twitter's audit and the Twitter ranking study found amplification of particular content. The studies measure different interventions on different platforms in different years.
| Study | Platform | Intervention | Length | Result on polarization |
|---|---|---|---|---|
| Guess et al., Science 2023 | Facebook, Instagram | Chronological feed | 3 months | No significant change |
| Guess et al., Science 2023 | Reshares removed | — | No significant effect | |
| Nyhan et al., Nature 2023 | Fewer like-minded sources | — | No measurable effect | |
| Allcott et al., PNAS 2024 | Facebook, Instagram | Deactivation | 6 weeks | Close to zero |
| Piccardi et al., Science 2025 | X | Reranked animosity content | 10 days | Out-party feelings shifted by more than 2 points of 100 |
The revenue facts in section 3.1 show how Meta earns money. They do not show that divisive content earned more than other content, and no filing separates revenue by content type. Facebook's stated position is that the 2018 change followed research on meaningful experiences and that advertisers avoid angry content.
A study that followed users over years, or that measured effects on content producers, would bear on the question. None of these studies did either.
3.12 Refunds at the two conduits
In the 2020 cycle WinRed reported $154.7 million in refunded individual contributions on $2.085 billion received, or 7.4 percent, and ActBlue reported $97.9 million refunded on $4.243 billion, or 2.3 percent.
Evidence: Exhibit 732
| Conduit | Cycle | Individual contributions | Refunded | Share refunded |
|---|---|---|---|---|
| WinRed | 2020 | $2,085.2 million | $154.7 million | 7.42% |
| WinRed | 2022 | $1,104.6 million | $74.4 million | 6.74% |
| WinRed | 2024 | $1,618.3 million | $67.1 million | 4.15% |
| ActBlue | 2016 | $679.9 million | $20.3 million | 2.98% |
| ActBlue | 2018 | $1,236.3 million | $38.6 million | 3.12% |
| ActBlue | 2020 | $4,243.1 million | $97.9 million | 2.31% |
| ActBlue | 2022 | $2,188.5 million | $55.5 million | 2.54% |
| ActBlue | 2024 | $3,758.6 million | $77.1 million | 2.05% |
| Committee | Individual contributions | Refunded |
|---|---|---|
| Donald J. Trump for President | $458.4 million | $11.5 million |
| Trump Make America Great Again Committee | $883.7 million | $0.02 million |
| Biden for President | $823.1 million | $15.8 million |
| Republican National Committee | $469.1 million | $0.68 million |
| Democratic National Committee | $231.7 million | $1.95 million |
WinRed began in 2019, so 2020 is its first full cycle. Its refund share fell in each cycle after 2020. The figures come from the FEC's OpenFEC committee totals, accessed 9 November 2026.
The FEC totals give no reason for a refund. Refunds occur for fraud, chargebacks, duplicate gifts, over-limit gifts and donor regret. The two conduits serve different donor bases and differ in size, and their receipts are pass-through amounts that also appear in recipient committees' totals. The ratio is not a measure of one party's tactics.
Exhibit 732 · Agency dataset
OpenFEC committee totals, 2016-2024 cycles: WinRed, ActBlue, campaign and party committees
FEC totals as filed: WinRed 2020 individual contributions $2,085.2M, refunded $154.7M; ActBlue 2020 $4,243.1M, refunded $97.9M. Also Trump, Biden, RNC, DNC 2020 committees. Refunds are not broken down by reason.
- Source
- Federal Election Commission (OpenFEC API)
- Date
- accessed 9 November 2026
- Grade
- Primary record
- Document
- https://api.open.fec.gov/v1/committee/C00694323/totals/
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.13 Regulators and a court
On 29 April 2021 the attorneys general of New York, Connecticut, Maryland and Minnesota wrote to WinRed about pre-checked recurring boxes, and on 6 May 2021 the FEC asked Congress to require affirmative consent for recurring contributions and to bar implied consent through pre-checked boxes. In 2023 the Eighth Circuit held in WinRed v. Ellison that federal election law did not block the Minnesota investigation.
Evidence: Exhibit 733, Exhibit 734, Exhibit 735, Exhibit 738
The FEC's 2021 legislative recommendations state that Commission staff are regularly contacted by individuals who have discovered recurring contributions charged to their credit cards or deducted from their checking accounts. The recommendations also ask for disclosures and for immediate cancellation on request.
The Eighth Circuit opinion of 7 February 2023 affirmed the dismissal of WinRed's suit. The court wrote that clear and conspicuous disclaimers about auto-recurring donations might have legitimated otherwise-illegal solicitation tactics, and it made no finding that WinRed or any campaign violated a law.
The opinion also recounts two earlier FEC matters. In MUR 7201, a single donor complained that the Ted Cruz for Senate committee charged two monthly gifts after she asked to cancel in September 2016. The Commission dismissed the matter 5 to 0 on prosecutorial discretion, noting that she had authorized the gifts and that they had been canceled. The Trump MAGA Committee matter, MUR 7255, closed without a violation of federal election law.
The official position
WinRed's blog post of 3 April 2021, a reply to the New York Times, says that campaigns decide how to use its tools, that it runs a refund and cancellation channel at all hours, and that ActBlue pioneered recurring and pre-checked boxes and continues to promote them.
What the record shows
The FEC, four state attorneys general and the Eighth Circuit are the primary record on pre-checked boxes. The record documents inquiries and a request for legislation. It contains no finding of a violation by WinRed.
Exhibit 733 · Agency report
Legislative Recommendations of the Federal Election Commission 2021
Commission staff are regularly contacted by individuals who have discovered recurring contributions to political committees have been charged to their credit card accounts or deducted from their checking accounts.
Asks Congress to require affirmative consent for recurring contributions, bar implied consent such as pre-checked boxes, require disclosures, and require immediate cancellation.
- Source
- Federal Election Commission
- Date
- 6 May 2021
- Quoted from
- p. 12, Explanation
- Grade
- Primary record
- Document
- https://www.fec.gov/resources/cms-content/documents/legrec2021.pdf
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 734 · Court opinion
WinRed, Inc. v. Ellison, 59 F.4th 934 (8th Cir. 2023), No. 22-1238
Clear and conspicuous disclaimers about auto-recurring donations might have legitimated otherwise-illegal solicitation tactics.
Affirms dismissal of WinRed's suit to block a Minnesota AG investigation of pre-checked recurring boxes. Recounts the 29 April 2021 joint letter from NY, CT, MD and MN AGs, the NYT report, and FEC matters MUR 7255 and MUR 7201. Does not find a violation.
- Source
- U.S. Court of Appeals for the Eighth Circuit
- Date
- 7 February 2023
- Quoted from
- Slip op. Part II (PDF p. 9)
- Grade
- Primary record
- Document
- https://ecf.ca8.uscourts.gov/opndir/23/02/221238P.pdf
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 735 · Agency enforcement file
MUR 7201 (Ted Cruz for Senate), Factual and Legal Analysis
The Complainant, who concedes she initially authorized the recurring contributions, does not appear to describe such a violation, and the Committee has confirmed that Complainant's recurring contributions have been canceled.
Single complainant alleged two monthly recurring contributions were charged after she asked to cancel in September 2016. Commission dismissed 5-0 on prosecutorial discretion; committee blamed a third-party vendor.
- Source
- Federal Election Commission
- Date
- 26 January 2018
- Quoted from
- Factual and Legal Analysis, p. 2
- Grade
- Primary record
- Document
- https://www.fec.gov/files/legal/murs/7201/18044436233.pdf
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
Exhibit 738 · Organization statement
The Facts the NYT Ignored
the use of which ActBlue pioneered years ago, and continues to promote today
WinRed's same-day response to the New York Times report. Says ActBlue pioneered recurring and pre-checked boxes, and that pre-checked boxes appeared on DCCC and DSCC ActBlue pages. Partisan self-description; use for WinRed's position only.
- Source
- WinRed
- Date
- 3 April 2021
- Quoted from
- Second paragraph
- Grade
- Primary record
- Document
- https://winred.com/blog/the-facts-the-nyt-ignored
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.14 The Republican-site experiment
A study of WinRed data on 14.8 million donations from 2.6 million donors estimated that weekly pre-checked recurring boxes, adopted by eight Republican campaign websites in 2020, increased donations by over $43 million and requested refunds by almost $3 million.
Evidence: Exhibit 736
Posner, Simonov, Mrkva and Johnson published the study in PNAS in 2023 using a staggered difference-in-differences design. The eight campaign sites in the sample are all Republican. The study says nothing about Democratic use or non-use of the same defaults. The authors note that the $3 million increase in requested refunds is small against $43 million in added donations.
The causal evidence on defaults therefore comes from one party's sites. Section 3.16 sets out what the record holds for the other.
Exhibit 736 · Peer-reviewed study
Dark defaults: How choice architecture steers political campaign donations
We estimate that defaults increased campaign donations by over $43 million while increasing requested refunds by almost $3 million.
Staggered difference-in-differences on 14.8 million donations by 2.6 million donors using WinRed FEC data; eight campaign websites, all Republican, adopted weekly pre-checked recurring boxes in 2020.
- Source
- Posner, Simonov, Mrkva, Johnson (PNAS 120)
- Date
- 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1073/pnas.2218385120
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.15 Emails from both parties
An analysis of more than 300,000 emails from over 3,000 senders in the 2020 cycle, including federal and state candidates and political committees across parties, found that manipulative tactics, meaning techniques that use some level of deception or clickbait, are the norm in political emails.
Evidence: Exhibit 737
Mathur and colleagues at Princeton's Center for Information Technology Policy published the study in Big Data and Society in January 2023. The corpus covers senders of all parties and several kinds of organization.
The study covers the 2020 cycle only.
Exhibit 737 · Peer-reviewed study
Manipulative tactics are the norm in political emails: Evidence from 300K emails from the 2020 US election cycle
manipulative tactics—techniques using some level of deception or clickbait—are the norm, not the exception.
Corpus of more than 300,000 emails from over 3,000 senders including federal and state candidates and PACs across parties.
- Source
- Mathur, Wang, Schwemmer, Hamin, Stewart, Narayanan (Big Data & Society)
- Date
- January 2023
- Quoted from
- Abstract
- Grade
- Peer-reviewed
- Document
- https://doi.org/10.1177/20539517221145371
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.16 The Democratic side of the record
WinRed's published account states that ActBlue pioneered recurring and pre-checked boxes and that pre-checked boxes appeared on ActBlue pages for the Democratic Congressional Campaign Committee and the Democratic Senatorial Campaign Committee. One press account reports that ActBlue began phasing them out in July 2021.
Evidence: Exhibit 738, Exhibit 734, Exhibit 739
WinRed made the first claim on 3 April 2021, the day the New York Times reported on the Trump campaign's recurring boxes. WinRed is a party to the dispute and its blog is a statement of its position. The state attorneys general letters to ActBlue are reported to have matched those sent to WinRed.
The sources for ActBlue's practices are WinRed's statement and press reports, and the register holds no ActBlue statement or attorneys general letter to ActBlue. The FEC totals in section 3.12 show ActBlue refunds of 2.3 percent in 2020, and the email study in section 3.15 covers Democratic senders.
Exhibit 739 · News report
How Trump Steered Supporters Into Unwitting Donations
The New York Times report of 3 April 2021 on the Trump campaign's pre-checked recurring boxes and refunds. The Eighth Circuit opinion in WinRed v. Ellison (exhibit 734) describes the report and press summaries describe it.
- Source
- The New York Times (Shane Goldmacher)
- Date
- 3 April 2021
- Grade
- Reported
- Document
- https://www.nytimes.com/2021/04/03/us/politics/trump-donations.html
- Accessed
- 2026-11-09
- Cited by
- Findings citing this exhibit
3.17 What the record leaves out
No cited document records an intent by a platform executive to produce division. Zuckerberg's and Rosen's statements deny it, and the internal documents are known only through reports.
No cited document records an intent by a campaign to charge donors who did not mean to give. The record on fundraising shows defaults, urgency and refunds, and campaigns' position is that donors choose recurring gifts and can cancel.
The record is thinnest in four places: the internal Facebook documents themselves, revenue by content type, ActBlue's and the Democratic committees' default practices, and refund rates for the party committees on both sides. The unsealed exhibits in the multistate attorneys general case against Meta and the FEC's itemized refund records would address the first and last.
Key findings
- Meta states that substantially all of its revenue is advertising, and its advertising revenue was $160.6 billion in 2024 (3.1).
- Facebook's 2017 annual report said the 2018 News Feed change would reduce time spent (3.2).
- Haugen told the Senate that Facebook's choices amplified division, and Facebook said it acted against polarization at a cost to revenue (3.3).
- Press reports of internal documents describe staff warnings and declined changes, and Facebook disputes them (3.4).
- Out-group posts were shared about twice as often, and social feedback increases outrage expression (3.5).
- Twitter's audit found right-leaning amplification in six of seven countries and no extreme-content premium (3.6).
- Twitter's engagement ranking amplified out-group hostile content (3.7).
- The 2020 Meta experiments found no significant change in polarization (3.8).
- A 2024 experiment on X shifted out-party feelings by more than 2 points (3.9).
- Two YouTube studies found little evidence of recommendation-driven radicalization (3.10).
- Whether engagement ranking causes lasting polarization is unresolved (3.11).
- WinRed refunded 7.4 percent and ActBlue 2.3 percent of 2020 individual contributions (3.12).
- The FEC, four attorneys general and the Eighth Circuit addressed pre-checked recurring boxes without a finding of violation (3.13).
- Pre-checked recurring boxes on eight Republican sites added over $43 million (3.14).
- Manipulative tactics are common in political emails from senders across parties (3.15).
- The Democratic-side record on defaults is reported and thinner (3.16).