Pith. sign in

REVIEW 3 major objections 4 minor 2 cited by

TikTok's recommendations skewed towards Republican content during the 2024 U.S. presidential race

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read During the 2024 U.S. presidential campaign, TikTok's recommendation algorithm served Republican-conditioned accounts about 11.8% more co-partisan videos than Democratic-conditioned accounts, and served Democratic-conditioned accounts…

desk verdict First controlled TikTok audit with a credible Republican skew, weakened but not sunk by transcript-only outcome measurement and narrow conditioning channel selection. read the letter →

arxiv 2501.17831 v2 pith:K7CVX2PP submitted 2025-01-29 cs.SI cs.CY

Hazem Ibrahim , HyunSeok Daniel Jang , Nouar Aldahoul , Aaron R. Kaufman , Talal Rahwan , Yasir Zaki This is my paper · ORCID
classification cs.SIcs.CY
keywords TikTokrecommendationalgorithmalgorithmicauditpartisanbias2024U.S.presidentialelectionsockpuppetaccountsnegativepartisanship
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to test whether TikTok's recommendation algorithm treats the two major U.S. parties symmetrically. Using 323 controlled 'sock puppet' accounts in New York, Texas, and Georgia, seeded with Democratic, Republican, or no partisan videos, the authors logged roughly 394,000 recommended videos from April to November 2024 and labeled them with an LLM ensemble. They find that Republican-seeded accounts received about 11.8% more party-aligned recommendations than Democratic-seeded accounts, while Democratic-seeded accounts received about 7.5% more opposite-party recommendations. The asymmetry appears in all three states and survives controls for video- and channel-level engagement, and it is driven mainly by negative-partisanship content, particularly anti-Democratic videos. If the finding holds, it would mean TikTok's opaque feed, on which users have only indirect control, systematically shapes political exposure differently for the two parties during a presidential election.

What carries the argument

The load-bearing mechanism is the sock-puppet algorithmic audit: freshly created TikTok accounts, geo-spoofed via VPN and GPS mocking, that first watch up to 400 researcher-selected partisan videos (the conditioning stage) and then log the videos appearing on the 'For You' page (the recommendation stage). The outcome measure is an ideological-content score ranging from -1 (all Democratic-aligned) to +1 (all Republican-aligned), computed as the proportion of Republican-aligned minus Democratic-aligned political videos. To rule out engagement-based explanations, the authors build 48 counterfactual models that sample videos proportional to observed video- and channel-level metrics, plus a sensitivity analysis showing that an unobserved engagement metric would need a Republican-Democrat gap roughly 9.4 times the best observed composite gap to explain the skew. A logistic-regression model of ideological mismatch, controlling for state, week, engagement, and conditioning, isolates the partisan conditioning effect.

What would settle it

A direct test would be an adversarial replication: fresh conditioning channels chosen by an independent panel, matched on content style and posting frequency, with all recommended videos (including those without transcripts) labeled by human annotators; if the Republican-conditioned co-partisan advantage disappears or reverses, the central claim fails. A complementary check would test whether any latent engagement feature available to TikTok has a Republican-Democrat gap larger than roughly 9.4 times the best observed composite gap.

Watch

Extended reading notes

Core claim

The central claim is that TikTok's recommendation algorithm displayed a measurable Republican skew during the 2024 U.S. presidential race. In paired weekly comparisons, Republican-conditioned bots were served approximately 11.8% more co-partisan videos than Democratic-conditioned bots, and Democratic-conditioned bots were served approximately 7.5% more cross-partisan videos. The authors operationalize this as an ideological-content score, the share of recommended political videos aligned with Republicans minus the share aligned with Democrats, and show the skew grew over the campaign and was present in a projected Democratic state, a projected Republican state, and a swing state. They argue the asymmetry is not a byproduct of engagement: counterfactual models that resample videos weighted by likes, comments, shares, plays, channel followers, or verification status predict little or no Republican advantage, and a logistic regression shows Democratic bots remain roughly 2.8 times as likely to receive ideologically mismatched videos even after controlling for comment partisanship. The mismatch is concentrated in negative-partisanship content and in top Republican channels such as the official Trump account and Fox News being recommended to Democratic bots.

Load-bearing premise

The central claim rests on the assumption that the 24 conditioning channels, matched only on follower count and cumulative likes, represent partisan political content on TikTok, and that stance labels inferred from transcripts (available for 22.8% of unique videos) faithfully represent the partisanship of all recommended videos.

Editorial extensions

If this is right

  • Democratic-leaning users can expect to see more opposing-party content than Republican-leaning users see, making the partisan experience asymmetric.
  • The Republican skew persists across New York, Texas, and Georgia, so it is not an artifact of one state's political environment.
  • Engagement metrics (likes, shares, comments, plays, followers, verification) do not explain the skew; counterfactual sampling by these metrics predicts little or no Republican advantage.
  • Negative-partisanship videos, especially Anti Democrat content, drive the mismatch; positive pro-party content contributes much less to the skew.
  • Top Republican channels, including Trump's and Fox News', were recommended to Democratic bots more often than top Democratic channels were recommended to Republican bots.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the skew is driven by engagement optimization around negative partisanship, then the same mechanism should produce out-party animosity biases in non-U.S. elections and in non-political domains where negativity drives watch time; this is a testable prediction the paper does not make.
  • A replication that matches conditioning channels on content format and posting frequency, rather than only on followers and likes, would reveal whether the asymmetry is about partisanship or about stylistic differences between the two channel sets.
  • Because stance labels come only from transcript-bearing videos (22.8% of unique videos), a multimodal replication that labels visual-only political content could shift the measured asymmetry; this is an open empirical question.
  • The survey of 1,000 users shows Republicans perceiving more co-partisan and positive content, consistent with the experimental skew, but linking those perceptions to the actual recommendation logs would be a stronger test of whether real users notice the asymmetry.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper reports a large-scale sock-puppet audit of TikTok's recommendation algorithm during the 2024 U.S. presidential race. The authors created 323 experimental runs across New York, Texas, and Georgia, conditioning fresh accounts with Democratic, Republican, or no partisan videos, and collected roughly 394,000 videos from April 30 to November 11, 2024. Political and partisan labels were assigned via a three-LLM ensemble with human validation on transcripts. The central finding is that Republican-conditioned bots received approximately 11.8% more co-partisan recommendations than Democratic-conditioned bots, while Democratic-conditioned bots received approximately 7.5% more cross-partisan recommendations. The authors argue that this asymmetry exists across all three states, persists after controlling for video- and channel-level engagement metrics, is driven mainly by negative partisanship content, and is not explained by observable engagement or asymmetric homophily. The paper also includes channel-level mismatch analyses, topic analyses, a misinformation exploration, and a survey of TikTok users' perceptions.

Significance. If the central result holds, the paper is a valuable empirical contribution to algorithmic auditing and to the public debate about platform neutrality. Its strengths include a longitudinal design spanning the election period, a large number of independent experimental runs, preregistration, a human-validated classification pipeline, and a serious set of robustness checks: counterfactual sampling under 48 engagement-based models, an asymmetric-homophily test, and a sensitivity analysis for latent engagement metrics. The replication materials and dataset are promised on GitHub, which is a further strength. However, the headline numbers are estimated only from videos with English transcripts, and the representativeness check does not test the between-condition difference that is the paper's main claim. This, together with the hand-picked conditioning channel set and unclustered video-level regressions, means the paper needs additional work before its central claim can be fully accepted.

major comments (3)
  1. [Methods, Data Representativeness; Measuring Political Content; Figure 7] The representativeness check in the Data Representativeness section tests whether transcript and non-transcript videos have the same marginal partisan distribution within each bot condition, but the paper's headline claim is a difference between conditions. The 11.8% and 7.5% estimates are computed only from the 40,264 transcript videos (22.8% of unique recommendations), and Figure 7 does not report an interaction test between bot condition and transcript availability. For example, if Republican-conditioned bots are disproportionately recommended Republican-aligned videos without transcripts, the transcript-only sample would understate or overstate the between-condition asymmetry. Please report the between-condition difference in partisan composition for the 4,000 human-annotated non-transcript videos, with a formal interaction test or a confidence interval, and state whether the headline asymmetry holds in that sample.
  2. [Methods, Conditioning Stage; Table 4] The conditioning treatment consists of 12 Democrat- and 12 Republican-aligned channels selected by searching politically charged keywords and matched on follower count and cumulative likes. This leaves open the possibility that unmeasured channel characteristics (content style, posting frequency, network position) are correlated with partisanship and influence downstream recommendations. A robustness analysis that reruns the conditioning with alternative or leave-one-out channel sets, or a model that includes channel-level random effects, would strengthen the claim that the asymmetry is attributable to partisan content rather than to the particular channels used. At minimum, the scope of the conclusion should be stated as conditional on the selected conditioning channels.
  3. [Figure 2H; Supplementary Tables 13-15] The logistic regressions are estimated at the bot-video level, but the text does not report clustering or a multilevel structure. Since videos are nested within bots and within channels, and each bot contributes up to 1,200 recommendations, the independence assumption is questionable. I request cluster-robust standard errors (by bot and/or by channel) or a mixed-effects specification, and a statement of whether the key coefficients (e.g., the Democrat bot odds ratio of 2.81 in Supplementary Table 13) remain significant after accounting for this dependence.
minor comments (4)
  1. [Introduction] The introduction states that the bots analyzed "over 340,000 videos encountered through 381 experimental runs," while the abstract reports roughly 394,000 videos and 323 experiments; these numbers should be reconciled.
  2. [Abstract and Introduction] The abstract reports 11.8% more co-partisan and 7.5% less cross-partisan content, while the introduction reports 11.5% and 8.0%; please clarify which definition corresponds to the headline estimate and use consistent numbers throughout.
  3. [Methods, Sensitivity Analysis] The sensitivity analysis refers to "Table 1 below" when it means Table 11, and the number of counterfactual models is inconsistent: the main text says 48 scenarios, the Methods says 39 models, and Supplementary Table 5 lists 54 rows; please correct the counts.
  4. [Robustness Checks, Sensitivity Analysis] The text says a latent metric would need to be "98 times bigger than the gap in likes" and also "9.4 times as strongly" as the combined metric; both statements are internally defensible, but they should be placed side by side or cross-referenced so readers do not mistake them for conflicting numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the headline asymmetry is a direct empirical measurement, with no free parameter fitted to produce it and no load-bearing self-citation.

full rationale

The central result—Republican-conditioned bots receiving approximately 11.8% more co-partisan and 7.5% less cross-partisan recommendations than Democratic-conditioned bots—is computed directly from observed recommendation outcomes conditioned by the experimental treatment. No parameter is fitted to make this difference appear, and the quantity is not defined in terms of itself: conditioning channels, LLM stance labels, and engagement counterfactuals are independent inputs. The engagement counterfactual models use observed video and channel metrics to generate expected skews that are then compared with, not fit to, the observed skew; the sensitivity analysis posits a hypothetical latent metric and asks how large it would have to be, which is an argument about magnitude rather than a fitted prediction. The paper's self-citations (e.g., its prior YouTube audit used for sock-puppet methodology) are methodological or background citations and are not load-bearing for the partisan-asymmetry claim. The representativeness check for the transcript-only sample concerns external validity (whether marginal partisan distributions generalize to non-transcript videos), which is a design limitation rather than a circular derivation. The claim is therefore self-contained as an empirical audit result.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests heavily on the representativeness of the 24 conditioning channels and on the validity of LLM-based stance labels over transcript-only videos. No invented entities or fitted free parameters are introduced; the main result is a direct comparison of measured proportions.

free parameters (1)
  • Channel alignment threshold = 75%
    Hand-set criterion for classifying channels as Democrat-aligned or Republican-aligned in the descriptive channel analysis. This threshold does not enter the main recommendation-rate outcome, which is computed from video-level LLM stance labels.
assumptions (4)
  • domain assumption The 24 conditioning channels (12 Democrat, 12 Republican) are representative of partisan political content on TikTok.
    The treatment is defined by watching videos from these specific channels. If these channels are atypical in algorithm-affecting features beyond followers and likes, the partisan asymmetry may not generalize. Matching was done on followers and total likes, but not on content style or network position.
  • domain assumption LLM majority-vote stance labels are valid proxies for video ideology.
    Validated on 500 human-annotated transcripts with majority accuracy 0.957 for the stance question, but the pipeline is not perfect and was applied only to videos with transcripts.
  • domain assumption Videos with transcripts are representative of all recommended videos for the partisan distribution of political content.
    Supported by human annotation of a 2,000-video sample per partisan bot group; the conditional partisan distribution matches, though the overall political rate differs. Still an assumption for the 77% of unique videos without transcripts.
  • domain assumption TikTok's algorithm treats bot watch behavior, consisting of 10-second views without likes or comments, as a valid signal of user preferences.
    The audit infers recommendation bias from bots that only watch videos. Real users' engagement and history may produce different recommendation mixes, and the external validity of sock-puppet audits is an accepted but untested premise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TikTok's recommendations skewed towards Republican content during the 2024 U.S. presidential race." pith.science (2026). https://pith.science/paper/K7CVX2PP

@misc{pith2026250117831,
  author       = {Pith},
  title        = {Pith review of: TikTok's recommendations skewed towards Republican content during the 2024 U.S. presidential race},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K7CVX2PP}},
  note         = {Machine review of arXiv:2501.17831}
}
read the original abstract

TikTok is a major force among social media platforms with over a billion monthly active users worldwide and 170 million in the United States. The platform's status as a key news source, particularly among younger demographics, raises concerns about its potential influence on politics in the U.S. and globally. Despite these concerns, there is scant research investigating TikTok's recommendation algorithm for political biases. We fill this gap by conducting 323 independent algorithmic audit experiments testing partisan content recommendations in the lead-up to the 2024 U.S. presidential elections. Specifically, we create hundreds of "sock puppet" TikTok accounts in Texas, New York, and Georgia, seeding them with varying partisan content and collecting algorithmic content recommendations for each of them. Collectively, these accounts viewed ~394,000 videos from April 30th to November 11th, 2024, which we label for political and partisan content. Our analysis reveals significant asymmetries in content distribution: Republican-seeded accounts received ~11.8% more party-aligned recommendations compared to their Democratic-seeded counterparts, and Democratic-seeded accounts were exposed to ~7.5% more opposite-party recommendations on average. These asymmetries exist across all three states and persist when accounting for video- and channel-level engagement metrics such as likes, views, shares, comments, and followers, and are driven primarily by negative partisanship content. Our findings provide insights into the inner workings of TikTok's recommendation algorithm during a critical election period, raising fundamental questions about platform neutrality.

Figures

Figures reproduced from arXiv: 2501.17831 by the authors.

Figure 1
Figure 1. Experimental Setup. 21 experiment runs are performed each week between April 30th and November 4th 2024. Bots are assigned an experimental condition based on two attributes, namely, their VPN and GPS location (New York, Texas, or Georgia), as well as the political leaning (Democratic or Republican) of the videos they watch during the conditioning stage. Each bot initially passes through a conditioning stage where it… view at source ↗
Figure 2
Figure 2. Differences in recommendation rates over time. (A-C) The mean ideological alignment for bots of a given conditioning-leaning in New York, Texas and Georgia, measured as the difference between the proportion of recommended videos which are Republican aligned minus the proportion of Democratic aligned videos. OLS regression lines are plotted with 95% confidence intervals. (D-F) The observed ideological skew for each s… view at source ↗
Figure 3
Figure 3. Mismatch proportions of top Democratic and Republican channels. (A) The mismatch pro￾portion, or the proportion of video watches by bots of an opposite conditioning-leaning for top Democratic and Republican TikTok channels by follower count. (B) The mean mismatch proportions of the TikTok accounts for the main political figures in the 2024 U.S. elections (Trump, JD Vance, Kamala Harris, and Tim Walz) (C) Logistic re… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Survey responses. Regression coefficients of Republican respondents relative to Democratic respondents for different survey questions regarding content on their TikTok feeds. The top four rows are with regards to responses to text-entry questions, while the bottom six …
Figure 6
Figure 6. Figure 6: Topic analysis of political TikTok videos. (A) The percentage of Republican-aligned videos on a given topic minus the percentage of Democratic-aligned videos on that topic. Statistically significant difference between Republican and Democratic percentages are highlight…
Figure 7
Figure 7. Figure 7: (A, B) A comparison between videos with and without transcripts seen by Democrat and Republican bots, respectively. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Algorithmic Audit of Personalisation Drift in Polarising Topics on TikTok

    cs.IR 2026-03 conditional novelty 6.0 of 10

    TikTok personalises strongly, neutralises climate/vaccine/flat-earth content toward safe neutral topics, but sustains and often stance-reinforces US politics—with a tilt toward the oppose stance when both sides are seeded.

  2. TikTok Rewards Divisive Political Messaging During the 2025 German Federal Election

    cs.SI 2025-09 conditional novelty 6.0 of 10

    Negative, angry, and outgroup-directed TikTok posts by German politicians attracted significantly more engagement than positive or relatable posts, and extreme parties both used and benefited from this style more than...

Reference graph

Works this paper leans on

92 extracted references · 76 canonical work pages · cited by 2 Pith papers

  1. [1]

    Crime generally819

    Crime818 1.1. Crime generally819

  2. [2]

    Climate change821 2.2

    Environment820 2.1. Climate change821 2.2. Other environmental issues822

  3. [3]

    Immigration generally824

    Immigration823 3.1. Immigration generally824

  4. [4]

    Abortion and reproductive health826 4.2

    Social issues825 4.1. Abortion and reproductive health826 4.2. Guns and gun control827 4.3. LGBTQ+ issues, including transgender issues828 4.4. Racial issues, including affirmative action and racial discrimination829 4.5. Education830 4.6. Other social issues, including culture war issues, labor, and other social issues that831 are not covered above832

  5. [5]

    Covid, including covid vaccines834 5.2

    Public health833 5.1. Covid, including covid vaccines834 5.2. Other vaccines835 5.3. Other public health issues836

  6. [6]

    Economy generally838

    Economy837 6.1. Economy generally838

  7. [7]

    AI, LLMs840 7.2

    Technology839 7.1. AI, LLMs840 7.2. Crypto841 7.3. Other technology issues842

  8. [8]

    Assassination attempt on Donald Trump844 36 8.2

    Government, politics and elections843 8.1. Assassination attempt on Donald Trump844 36 8.2. Republican National Convention (RNC) 845 8.3. Democratic National Convention (DNC) 846 8.4. Biden dropping out of the presidential race 847 8.5. Other political or government related posts that do not fit into other categories 848

Show all 92 references
  1. [9]

    Israel, Gaza or Palestine, including anything about Netanyahu or Hamas 850 9.2

    International issues 849 9.1. Israel, Gaza or Palestine, including anything about Netanyahu or Hamas 850 9.2. Ukraine war 851 9.3. Anything outside the US or involve US foreign relations except for Israel, Gaza, 852 Ukraine, or immigration 853

  2. [10]

    quality” while isolating the effect of the choices made by the algorithm, we normalize these metrics 896 as follows. First, we take the number of “recommendations

    No topic 854 10.1. None of the above topics 855 Using the above prompt, the topics discussed in the 7,767 ideologically stanced TikTok videos 856 in our dataset were identified. Table 13 below describes the number of videos falling under each 857 category, split by the video’s...

  3. [11]

    TikTok Statistics

    Doyle, B. TikTok Statistics. https://wallaroomedia.com/blog/social-media/tiktok-952 statistics/. [Accessed 28-11-2024].953

    2024

  4. [12]

    https://www.bloomberg.com/954 news/articles/2024-01-31/tiktok-ceo-pledges-2-billion-to-protect-170-million-955 us-users

    TikTok CEO pledges 2 billion to protect 170 million US users. https://www.bloomberg.com/954 news/articles/2024-01-31/tiktok-ceo-pledges-2-billion-to-protect-170-million-955 us-users. [Accessed 28-11-2024].956

    2024

  5. [13]

    6 facts about Americans and TikTok — pewresearch.org.https://www.pewresearch.957 org/short-reads/2024/04/03/6-facts-about-americans-and-tiktok/

    Eddy, K. 6 facts about Americans and TikTok — pewresearch.org.https://www.pewresearch.957 org/short-reads/2024/04/03/6-facts-about-americans-and-tiktok/ . [Accessed 28-11-958 2024].959

    2024

  6. [14]

    & Matsa, K

    Leppert, R. & Matsa, K. E. More Americans – especially young adults – are regularly getting960 news on TikTok — pewresearch.org. https://www.pewresearch.org/short-reads/2024/09/961 17/more-americans-regularly-get-news-on-tiktok-especially-young-adults/ . [Ac-962 cessed 28-11-2024].963

    2024

  7. [15]

    TikTok. Politics, Religion, and Culture — TikTok Advertising Policies — ads.tiktok.com.964 https://ads.tiktok.com/help/article/tiktok-ads-policy-politics-religion-and-965 culture (October, 2024). [Accessed 28-11-2024].966

    2024

  8. [16]

    & Haeck, P

    Vinocur, N. & Haeck, P. TikTok CEO summoned to European Parliament over role in shock967 Romania election — politico.eu. https://www.politico.eu/article/elections-tiktok-968 ceo-eu-parliament-romania-election-fake-accounts-pro-russia-calin-georgescu-969 nato-shock-victory/. [A...

    2024

  9. [17]

    Romania regulator calls for TikTok suspension amid vote interference fears — the-971 guardian.com

    Henly, J. Romania regulator calls for TikTok suspension amid vote interference fears — the-971 guardian.com. https://www.theguardian.com/world/2024/nov/27/romanian-regulator-972 tiktok-suspended-cyber-interference-election-georgescu . [Accessed 28-11-2024].973

    2024

  10. [18]

    The EU’s Digital Services Act — commission.europa.eu

    Commission, E. The EU’s Digital Services Act — commission.europa.eu. https:974 //commission.europa.eu/strategy-and-policy/priorities-2019-2024/europe-fit-975 digital-age/digital-services-act_en . [Accessed 28-11-2024].976

    2019

  11. [19]

    House, T. W. Addressing the Threat Posed by TikTok, and Taking Additional Steps to Address977 the National Emergency With Respect to the Information and Communications Technology and978 Services Supply Chain. https://www.federalregister.gov/documents/2020/08/11/2020-979 17699/...

    2020

  12. [20]

    Allyn, B. U.S. Judge Halts Trump’s TikTok Ban, The 2nd Court To Fully Block The 982 Action. https://www.npr.org/2020/12/07/944039053/u-s-judge-halts-trumps-tiktok- 983 ban-the-2nd-court-to-fully-block-the-action (2020). [Accessed 28-11-2024]. 984

    2020

  13. [21]

    Jeff Stein, D. H. & Bogage, J. Trump expected to try to halt TikTok ban, allies say. https:// 985 www.washingtonpost.com/business/2024/11/12/trump-tiktok-ban-sale/ . [Accessed 28- 986 11-2024]. 987

    2024

  14. [22]

    & McCabe, D

    Maheshwari, S. & McCabe, D. Congress Passed a Bill That Could Ban TikTok. Now 988 Comes the Hard Part. — nytimes.com. https://www.nytimes.com/2024/04/23/technology/ 989 bytedance-tiktok-ban-bill.html . [Accessed 28-11-2024]. 990

    2024

  15. [23]

    Jack Ma loses more than half of his wealth after criticizing Chinese regulators 991 — CNN Business — edition.cnn.com

    Toh, M. Jack Ma loses more than half of his wealth after criticizing Chinese regulators 991 — CNN Business — edition.cnn.com. https://edition.cnn.com/2023/07/12/business/ 992 china-jack-ma-wealth-drop-intl-hnk/index.html . [Accessed 29-11-2024]. 993

    2023

  16. [24]

    China announces rules to reduce spending on video games

    Ye, J. China announces rules to reduce spending on video games. https://www.reuters.com/ 994 world/china/china-issues-draft-rules-online-game-management-2023-12-22/ . [Ac- 995 cessed 29-11-2024]. 996

    2023

  17. [25]

    Revealed: how TikTok censors videos that do not please Beijing — the- 997 guardian.com

    Hern, A. Revealed: how TikTok censors videos that do not please Beijing — the- 997 guardian.com. https://www.theguardian.com/technology/2019/sep/25/revealed-how- 998 tiktok-censors-videos-that-do-not-please-beijing . [Accessed 29-11-2024]. 999

    2019

  18. [26]

    TikTok apologized for the glitch affecting the ’black lives matter’ hashtag 1000 after accusations of censorship: ’We know this came at a painful time’ — businessin- 1001 sider.com

    Harris, M. TikTok apologized for the glitch affecting the ’black lives matter’ hashtag 1000 after accusations of censorship: ’We know this came at a painful time’ — businessin- 1001 sider.com. https://www.businessinsider.com/tiktok-apologizes-for-blm-hashtag- 1002 glitch-after...

    2020

  19. [27]

    Trump likely to give tiktok a 90-day extension to avoid ban (2025)

    News, N. Trump likely to give tiktok a 90-day extension to avoid ban (2025). URL 1004 https://www.nbcnews.com/politics/donald-trump/trump-likely-give-tiktok-90-day- 1005 extension-avoid-ban-rcna188258 . [Accessed: 2025-01-19]. 1006

    2025

  20. [28]

    Guess, A. M. et al. Reshares on social media amplify political news but do not detectably affect 1007 beliefs or opinions. Science 381, 404–408 (2023). 1008

    2023

  21. [29]

    Asymmetric ideological segregation in exposure to political news on 1009 facebook

    Gonz´ alez-Bail´ on, S.et al. Asymmetric ideological segregation in exposure to political news on 1009 facebook. Science 381, 392–398 (2023). 1010

    2023

  22. [30]

    Nyhan, B. et al. Like-minded sources on facebook are prevalent but not polarizing. Nature 1011 620, 137–144 (2023). 1012 43

    2023

  23. [31]

    Facebooking it to the polls: A study in online social networking and political behavior.1013 Journal of Information Technology & Politics 9, 352–369 (2012).1014

    Bode, L. Facebooking it to the polls: A study in online social networking and political behavior.1013 Journal of Information Technology & Politics 9, 352–369 (2012).1014

    2012

  24. [32]

    & Grzywi´ nska, I

    Batorski, D. & Grzywi´ nska, I. Three dimensions of the public sphere on facebook.Information,1015 Communication & Society 21, 356–374 (2018).1016

    2018

  25. [33]

    & Adamic, L

    Bakshy, E., Messing, S. & Adamic, L. A. Exposure to ideologically diverse news and opinion1017 on facebook. Science 348, 1130–1132 (2015).1018

    2015

  26. [34]

    & Sunstein, C

    Quattrociocchi, W., Scala, A. & Sunstein, C. R. Echo chambers on facebook. Available at1019 SSRN 2795110 (2016).1020

    2016

  27. [35]

    Allcott, H. et al. The effects of facebook and instagram on the 2020 election: A deactivation1021 experiment. Proceedings of the National Academy of Sciences 121, e2321584121 (2024).1022

    2024

  28. [36]

    & Raynauld, V

    Lalancette, M. & Raynauld, V. The power of political image: Justin trudeau, instagram, and1023 celebrity politics. American behavioral scientist 63, 888–924 (2019).1024

    2019

  29. [37]

    Trevisan, M. et al. Towards understanding political interactions on instagram. In Proceedings1025 of the 30th ACM conference on hypertext and social media , 247–251 (2019).1026

    2019

  30. [38]

    Parmelee, J. H. & Roman, N. Insta-politicos: Motivations for following political leaders on1027 instagram. Social media+ society 5, 2056305119837662 (2019).1028

    2019

  31. [39]

    & Zaki, Y

    Ibrahim, H., AlDahoul, N., Lee, S., Rahwan, T. & Zaki, Y. Youtube’s recommendation algo-1029 rithm is left-leaning in the united states. PNAS nexus 2, pgad264 (2023).1030

    2023

  32. [40]

    Hosseinmardi, H. et al. Evaluating the scale, growth, and origins of right-wing echo chambers1031 on youtube. arXiv preprint arXiv:2011.12843 (2020).1032

    2020 arXiv

  33. [41]

    Hosseinmardi, H. et al. Examining the consumption of radical content on youtube. Proceedings1033 of the National Academy of Sciences 118, e2101967118 (2021).1034

    2021

  34. [42]

    & Zaitsev, A

    Ledwich, M. & Zaitsev, A. Algorithmic extremism: Examining youtube’s rabbit hole of radi-1035 calization. arXiv preprint arXiv:1912.11211 (2019).1036

    2019 arXiv

  35. [43]

    Lambrecht, A., Sen, A., Tucker, C. E. & Wiertz, C. Algorithmic recommendations and earned1037 media: Investigating product echo chambers on youtube. Available at SSRN 3951425 (2021).1038

    2021

  36. [44]

    Balasubramanian, A. et al. A public dataset tracking social media discourse about the 2024 us1039 presidential election on twitter/x. arXiv preprint arXiv:2411.00376 (2024).1040

    2024 arXiv

  37. [45]

    & Arvidsson, A

    Colleoni, E., Rozza, A. & Arvidsson, A. Echo chamber or public sphere? predicting political ori-1041 entation and measuring political homophily in twitter using big data.Journal of communication1042 64, 317–332 (2014).1043 44

    2014

  38. [46]

    Conover, M. et al. Political polarization on twitter. In Proceedings of the international aaai 1044 conference on web and social media , vol. 5, 89–96 (2011). 1045

    2011

  39. [47]

    T., Nagler, J., Tucker, J

    Barber´ a, P., Jost, J. T., Nagler, J., Tucker, J. A. & Bonneau, R. Tweeting from left to right: 1046 Is online political communication more than an echo chamber? Psychological science 26, 1047 1531–1542 (2015). 1048

    2015

  40. [48]

    E., Kaltenbrunner, A., Laniado, D

    Arag´ on, P., Kappler, K. E., Kaltenbrunner, A., Laniado, D. & Volkovich, Y. Communication 1049 dynamics in twitter during political campaigns: The case of the 2011 spanish national election. 1050 Policy & internet 5, 183–206 (2013). 1051

    2013

  41. [49]

    Ibrahim, H. et al. Analyzing political stances on twitter in the lead-up to the 2024 us election. 1052 arXiv preprint arXiv:2412.02712 (2024). 1053

    2024 arXiv

  42. [50]

    & Starnini, M

    Cinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W. & Starnini, M. The 1054 echo chamber effect on social media. Proceedings of the National Academy of Sciences 118, 1055 e2023301118 (2021). 1056

    2021

  43. [51]

    & Rossetti, G

    Morini, V., Pollacci, L. & Rossetti, G. Toward a standard approach for echo chamber detection: 1057 reddit case study. Applied Sciences 11, 5390 (2021). 1058

    2021

  44. [52]

    & Starnini, M

    De Francisci Morales, G., Monti, C. & Starnini, M. No echo in the chambers of political 1059 interactions on reddit. Scientific Reports 11, 2818 (2021). 1060

    2021

  45. [53]

    C., Papakyriakopoulos, O

    Medina Serrano, J. C., Papakyriakopoulos, O. & Hegelich, S. Dancing to the partisan beat: A 1061 first analysis of political communication on tiktok. In Proceedings of the 12th ACM Conference 1062 on Web Science , 257–266 (2020). 1063

    2020

  46. [54]

    & Mar ´ ın-Llad´ o, C

    Cervi, L. & Mar ´ ın-Llad´ o, C. What are political parties doing on tiktok? the spanish case. 1064 Profesional de la informaci´ on30 (2021). 1065

    2021

  47. [55]

    & Fahmy, S

    Zamora-Medina, R., Suminas, A. & Fahmy, S. S. Securing the youth vote: A comparative 1066 analysis of digital persuasion on tiktok among political actors. Media and communication 11, 1067 218–231 (2023). 1068

    2023

  48. [56]

    The use of tiktok for political campaigning in canada: the case of jagmeet singh

    Moir, A. The use of tiktok for political campaigning in canada: the case of jagmeet singh. 1069 Social Media+ Society 9, 20563051231157604 (2023). 1070

    2023

  49. [57]

    & Votta, F

    Guinaudeau, B., Munger, K. & Votta, F. Fifteen seconds of fame: Tiktok and the supply side 1071 of social video. Computational Communication Research 4, 463–485 (2022). 1072

    2022

  50. [58]

    Vasconcellos, P. H. S., Lara, P. D. d. A. & Marques-Neto, H. T. Analyzing polarization and 1073 toxicity on political debate in brazilian tiktok videos transcriptions. In Proceedings of the 15th 1074 45 ACM Web Science Conference 2023, WebSci ’23, 33–42 (Association for Comput...

    2023

  51. [59]

    & Abidin, C

    Zeng, J. & Abidin, C. ‘# okboomer, time to meet the zoomers’: Studying the memefication of1077 intergenerational politics on tiktok. In The playful politics of memes , 93–115 (Routledge, 2023).1078

    2023

  52. [60]

    & Pavlidis, A

    Brown, Y., Pini, B. & Pavlidis, A. Affective design and memetic qualities: Generating affect1079 and political engagement through bushfire tiktoks. Journal of sociology 60, 121–137 (2024).1080

    2024

  53. [61]

    Ackerman, B. A. What is neutral about neutrality? Ethics 93, 372–390 (1983).1081

    1983

  54. [62]

    McCarthy, K. J. & Dolfsma, W. Neutral media? evidence of media bias and its economic1082 impact. Review of Social Economy 72, 42–54 (2014).1083

    2014

  55. [63]

    Hello gpt-4 turbo

    OpenAI. Hello gpt-4 turbo. https://openai.com/index/hello-gpt-4o/ (2025). Accessed:1084 2025-01-27.1085

    2025

  56. [64]

    Gemini pro

    DeepMind. Gemini pro. https://deepmind.google/technologies/gemini/pro/ (2025). Ac-1086 cessed: 2025-01-27.1087

    2025

  57. [65]

    OpenAI. Gpt-4. https://openai.com/index/gpt-4/ (2025). Accessed: 2025-01-27.1088

    2025

  58. [66]

    Tjaden, J. et al. Automated election audits-analyzing exposure to political content on social1089 media with a case study of tiktok in germany’s 2024 regional elections. OSF (2024).1090

    2024

  59. [67]

    Tikapi: Tiktok api for developers (2025)

    TikAPI. Tikapi: Tiktok api for developers (2025). URL https://tikapi.io/.1091

    2025

  60. [68]

    Abramowitz, A. I. & Webster, S. W. Negative partisanship: Why americans dislike parties but1092 behave like rabid partisans. Political Psychology 39, 119–135 (2018).1093

    2018

  61. [69]

    America’s News Influencers — pewresearch.org

    Stocking, G. America’s News Influencers — pewresearch.org. https://www.pewresearch.org/1094 journalism/2024/11/18/americas-news-influencers/ . [Accessed 02-01-2025].1095

    2024

  62. [70]

    Snopes.com — snopes.com

    Snopes. Snopes.com — snopes.com. snopes.com. [Accessed 28-01-2025].1096

    2025

  63. [71]

    PolitiFact — politifact.com

    Politifact. PolitiFact — politifact.com. politifact.com. [Accessed 28-01-2025].1097

    2025

  64. [72]

    [Accessed 13-01-2025].1100

    Young Voters Shifted Toward Trump but Still Favored Harris Overall — now.tufts.edu.1098 https://now.tufts.edu/2024/11/12/young-voters-shifted-toward-trump-still-1099 favored-harris-overall. [Accessed 13-01-2025].1100

    2024

  65. [73]

    Ziems, C. et al. Can large language models transform computational social science? Computa-1101 tional Linguistics 50, 237–291 (2024).1102

    2024

  66. [74]

    & Tesconi, M

    Gambini, M., Senette, C., Fagni, T. & Tesconi, M. Evaluating large language models for user1103 stance detection on x (twitter). Machine Learning 113, 7243–7266 (2024).1104 46

    2024

  67. [75]

    Bruna Horvath, J. A. & Goggin, B. Meta is ending its fact-checking program in favor of a 1105 ’community notes’ system similar to X’s — nbcnews.com. https://www.nbcnews.com/news/ 1106 amp/rcna186468 (07-01-2025). [Accessed 08-01-2025]. 1107

    2025

  68. [76]

    & Langbort, C

    Sandvig, C., Hamilton, K., Karahalios, K. & Langbort, C. Auditing algorithms: Research 1108 methods for detecting discrimination on internet platforms. Data and discrimination: convert- 1109 ing critical concerns into productive inquiry 22, 4349–4357 (2014). 1110

    2014

  69. [77]

    Haroon, M. et al. Auditing youtube’s recommendation system for ideologically congenial, ex- 1111 treme, and problematic recommendations. Proceedings of the National Academy of Sciences 1112 120, e2213020120 (2023). 1113

    2023

  70. [78]

    Web scraping social media: Pitfalls of copyright and data protection 1114 law (2020)

    Golla, S. Web scraping social media: Pitfalls of copyright and data protection 1114 law (2020). URL https://blog.prif.org/2020/04/15/web-scraping-social-media- 1115 pitfalls-of-copyright-and-data-protection-law/ . Accessed: 2025-01-07. 1116

    2020

  71. [79]

    X (formerly twitter) and elon musk sued by hate speech watchdog in free speech 1117 case (2024)

    Press, A. X (formerly twitter) and elon musk sued by hate speech watchdog in free speech 1117 case (2024). URL https://apnews.com/article/x-twitter-musk-hate-speech-lawsuit- 1118 fafa1904f5525f9ab64250e81a72d210. Accessed: 2025-01-07. 1119

    2024

  72. [80]

    Top names of the 2010s (2025)

    Social Security Administration. Top names of the 2010s (2025). URL https://www.ssa.gov/ 1120 oact/babynames/decades/names2010s.html. Accessed: 2025-01-24. 1121

    2025

  73. [81]

    Location services on tiktok (2024)

    TikTok Support. Location services on tiktok (2024). URL https://support.tiktok.com/en/ 1122 account-and-privacy/account-privacy-settings/location-services-on-tiktok . Ac- 1123 cessed: 2024-12-13. 1124

    2024

  74. [82]

    imyfone location changer (2024)

    iMyFone. imyfone location changer (2024). URL https://www. 1125 imyfone.com/location-changer/?media=googleppc&gad_source=1&gclid= 1126 EAIaIQobChMI1en4tJGkigMVJrKDBx1SpRS2EAAYASABEgIY9vD_BwE. Accessed: 2024-12-13. 1127

    2024

  75. [83]

    Wi-fi mac randomization behavior in android (2025)

    Developers, A. Wi-fi mac randomization behavior in android (2025). URL 1128 https://source.android.com/docs/core/connect/wifi-mac-randomization-behavior#: 1129 ~:text=Android%20generates%20a%20persistent%20randomized,the%20same%20until% 1130 20factory%20reset. Accessed: 2025-0...

    2025

  76. [84]

    Vector embeddings

    OpenAI. Vector embeddings. https://platform.openai.com/docs/guides/embeddings# 1132 embedding-models (2024). [Accessed 28-01-2025]. 1133

    2024

  77. [85]

    & Aral, S

    Vosoughi, S., Roy, D. & Aral, S. The spread of true and false news online. science 359, 1134 1146–1151 (2018). 1135 47 Survey Materials1136 Informed Consent1137 You are being asked to provide consent to participate in a research study. Participation is1138 voluntary. You can s...

    2018

  78. [86]

    Your participation is voluntary.1159

  79. [87]

    You are 18 years of age or older.1160

  80. [88]

    You may withdraw at any time.1161 Options:1162 • I consent, begin the study.1163 • I do not consent, I do not wish to participate.1164 48 Survey Questions 1165 Free Response Questions 1166 Q1) Over the past 12 months, from March 2024 to today, how has the overall content of yo...

    2024

  81. [89]

    Pre-Experiment Protocol

  82. [90]

    24-Hour Sleep (a) Randomly shuffle 8 conditioning channels (b) Watch ~50 videos from each channel

  83. [91]

    Recommendation Stage (a) Open TikTok and Watch 10 videos on For You (b) Quit TikTok and Sleep 1 Hour (c) Repeat

  84. [92]

    Post-Experiment Protocol Channel #5 Channel #3 Channel #7 Channel #4 ... Shuffle . . . x10 Sleep (1 Hour) Channel #1 Channel #2 Channel #3 Channel #8 ... Channel #5 x50 Channel #3 x50 Channel #7 x50 Channel #4 x50 . . . Watch Watch (a) Set GPS & IP to NY/TX/GA (b) Install TikTo...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.