TikTok’s Recommendation System Shows Partisan Skew in 2024 Election Content

A team from New York University Abu Dhabi’s AI and Society Lab created 323 automated “bot” accounts to audit TikTok’s For You feed during the 2024 U.S. election. Between 30 April and 11 November 2024, the bots—aged 22-24 and assigned to New York, Texas, and Georgia—watched up to 400 videos from either Republican or Democratic creators, while a neutral control group in Georgia received no political training. Over 27 weeks the experiment collected more than 280 000 recommended videos, of which 40 264 had downloadable transcripts for analysis.

Psypost

The study addresses a long-standing debate over whether online polarization stems from user self-selection or algorithmic amplification. TikTok’s reliance on a single recommendation feed makes it a “clean” environment for isolating algorithmic effects, especially as political content surged during the election cycle. Researchers used location-masking software, factory-reset phones each week, and an ensemble of three AI language models—validated by political-science students—to classify videos as political, election-related, and partisan.

Psypost

The work was authored by Hazem Ibrahim, HyunSeok Daniel Jang, Nouar Aldahoul, Aaron R. Kaufman, Talal Rahwan (corresponding author), and Yasir Zaki. All are affiliated with NYU Abu Dhabi’s AI and Society Lab.

Psypost

• Republican-trained bots received ~11.5 % more content aligned with their own party than Democratic-trained bots.

Psypost

• Democratic-trained bots were exposed to ~7.5 % more cross-party content.

Psypost

• The asymmetry was driven primarily by anti-Democratic videos appearing on Democratic-leaning feeds, not by a generic increase in Republican material.

Psypost

• Skewed content clustered in immigration, crime, and foreign-policy topics for Democrats, and abortion for Republicans.

Psypost

• A supplemental survey of 1 008 U.S. TikTok users (mostly 25-34 year-olds, slightly Republican-skewed) found conservatives more likely to report seeing optimistic, pro-Trump posts, while Democrats reported more anti-Democratic material.

Psypost

Researchers emphasized that the findings document a consistent exposure pattern across states and user types, without attributing intent to TikTok. They clarified that the study measures exposure, not persuasion, and that the bots simulate new users with short engagement histories, which may differ from long-tenured users. Limitations noted include the focus on English-language videos and the possibility that creator supply dynamics, rather than algorithmic bias, contributed to the observed skew.

Psypost

The authors themselves highlighted several constraints:

Psypost

• The dataset excludes Spanish-language and other minority-language content, leaving a gap in understanding for non-English speakers.

Psypost

• Exposure does not equate to attitude change, so causal effects on voting behavior remain untested.

Psypost

• Short-term bot behavior may not reflect the experience of established users with richer interaction histories.

Psypost

• Survey responses were drawn from a sample that leaned Republican, potentially biasing self-reported perceptions of feed tone.

Psypost

• The study’s reliance on transcript analysis omits visual and audio cues that could influence political messaging.

Psypost

• No direct evidence was provided regarding TikTok’s internal policy decisions or algorithmic design choices.

Psypost

• “TikTok’s feed isn’t a neutral window into politics,” — Yasir Zaki, Researcher

Psypost

• “The platform’s recommendations treat Democrats and Republicans differently, consistently, across states, and in ways that can’t be explained by differences in how people engage with the content.” — Talal Rahwan, Corresponding Author

Psypost

• “We are not saying TikTok deliberately chose to favor Republicans, our study documents a pattern in outcomes, not intent,” — Yasir Zaki, Researcher

Psypost

• “That said, we measure exposure, not persuasion, so we can’t say this changed anyone’s vote,” — Talal Rahwan, Corresponding Author

Psypost

• “The gaps are averages across hundreds of experiments over six months that held up across three states and survived 48 robustness checks, on a platform serving political content to tens of millions of young voters daily, that consistency matters,” — Talal Rahwan, Corresponding Author

Psypost

• “We’d like to combine bot audits with real user data, develop methods to capture visual and audio political messaging beyond transcripts, and run cross-platform comparisons,” — Hazem Ibrahim, Researcher

Psypost

The study suggests that TikTok’s recommendation engine can produce systematic partisan exposure, raising questions about the platform’s role in shaping political information for young voters. The research team plans to extend audits to real-user data, incorporate multilingual content, and compare algorithmic behavior across social-media platforms to assess broader democratic impacts.

Psypost