Recent research has shown evidence that TikTok’s recommendation system tends to expose users to more conservative and anti-Democratic political content than liberal content. This ideological imbalance occurs regardless of users’ initial political interests, suggesting that automation plays an important role in modern information access. The study was published in a scientific journal nature.
Researchers at New York University Abu Dhabi’s Institute for AI and Society conducted a new study exploring how automated internet systems are shaping the political news people see. Experts debate whether the internet’s polarization is caused by people seeking their preferred viewpoints, or by computer algorithms pushing out extreme content to keep users interested.
Corresponding author Talal Rahwan pointed out that this question is difficult to answer on older platforms. “In our previous work, we investigated recommendation algorithms on YouTube, but disentangling the algorithm’s influence on user self-selection has been a persistent challenge,” he told PsyPost.
TikTok provides a conducive environment for testing the impact of algorithms, as its primary interface relies heavily on automation. “TikTok’s For You page provided a uniquely clean setting to study that question, especially as political content on the platform becomes increasingly important in the 2024 election, as algorithms drive nearly everything users see,” Rahwan added.
To collect the data, the researchers created 323 automated bot accounts that acted as artificial users. This method is also known as sockpuppet auditing. From April 30 to November 11, 2024, the scientists launched 21 new accounts every week. To mimic the habits of young adult voters, each bot was assigned an age between 22 and 24 years.
The research team used location masking software to virtually place these phones in specific areas of the country. They assigned bots to New York, which represents areas that are reliably Democratic, Texas, which represents areas that are reliably Republican, and Georgia, which serves as a battleground state. The scientists used physical smartphones running Android software and reset the smartphones to factory settings after weekly tests. This process prevented the app from tracking devices across different weeks.
The account first went through a training phase to teach the app its assumed political preferences. Some bots watched up to 400 videos from known Republican creators, while others watched up to 400 videos from Democratic creators. The bot watched each of these training videos for exactly one minute. In Georgia, we skipped this political training phase entirely and also included a set of neutral bots that served as a control group of politically apathetic users.
After the training phase, the account switched to the main feed to see what the algorithm recommended. The bot watched the first 10 seconds of each recommended video before moving on to the next video. Researchers collected more than 280,000 recommended videos across 27 weeks. They were able to download the text transcripts of 40,264 of these videos, providing a large dataset of audio content for analysis.
The authors used an ensemble of artificial intelligence language models to classify the political nature of transcriptions. They combined the outputs of three different models to prevent their results from being skewed by a single program. Human and political science students verified the accuracy of these automated programs. The tool determined whether a video was political, related to the upcoming election, and supported or opposed a particular political party.
This finding suggests a clear asymmetry in the way political content is distributed on platforms. Accounts trained on Republican videos received approximately 11.5% more content aligned with their party compared to Democratic accounts. At the same time, Democratic accounts were exposed to about 7.5% more nonpartisan content.
Co-author Yasir Zaki explained the broader implications of these numbers. “The TikTok feed is not a neutral window into politics,” he said. “The platform’s recommendations consistently treat Democrats and Republicans differently across states and in ways that cannot be explained by differences in how people engage with the content.”
Hazem Ibrahim, also a researcher at New York University Abu Dhabi, pointed out the unusual nature of these results. “The bias was specifically centered around anti-Democratic content being pushed to Democratic-leaning accounts, rather than a general ‘more Republican content’ effect,” Ibrahim said.
The topics of these recommended videos were also broadly divided. “We were also struck by how the asymmetry was not spread evenly across the political spectrum, but concentrated in specific policy areas, particularly immigration, crime, and foreign policy for Democrats and abortion for Republicans,” Ibrahim added.
To see if real people were aware of these trends, researchers surveyed 1,008 active TikTok users in the United States online. The survey asked participants whether they had noticed a change in the political tone of their feeds over the past year. The online sample skewed slightly toward Republicans and consisted primarily of adults between the ages of 25 and 34.
Survey responses provided evidence that real-world experiences matched automated bot experiments. Republican respondents were significantly more likely than Democratic respondents to report seeing positive political content that agreed with their views. When asked open-ended questions, conservative users frequently mentioned an increase in optimistic and pro-Trump posts in their daily feeds.
Rahan emphasized the rigorous statistical testing behind the findings. “This gap is the average of hundreds of experiments over six months, maintained in three states and surviving 48 robustness checks, on a platform that provides political content to tens of millions of young voters every day, where consistency is key,” Rahwan said.
Robustness checks are mathematical tests used to ensure that results are correct under different conditions. Still, Rawan pointed out that exposure does not equate to influence. “However, we are measuring exposure, not persuasion, so we cannot say that this changed anyone’s vote,” he added.
Zaki also stressed that the investigation does not prove that the company intentionally programmed its software to favor conservatives. “We’re not saying that TikTok intentionally chose to support Republicans; our study documents a pattern of outcomes, not intentions,” Zaki said.
Auto Accounts were specifically designed to represent new users interacting with the platform for the first time. “We also study content exposure, rather than attitude change. Our bot simulates new users with a short engagement history, so long-time users may have a different experience,” Zaki said.
This difference may also be caused by fundamental supply differences in the types of videos that political creators choose to upload during election periods. Additionally, this study only examined videos in English. This excludes the experiences of Spanish speakers and other minority language communities in the United States.
The researchers plan to expand this work through a research group that can be viewed online at ai-and-society.com. “We want to combine bot audits with real user data, develop ways to capture visual and audio political messages beyond transcripts, and perform cross-platform comparisons,” Ibrahim said. “A major unanswered question is linking these exposure patterns to downstream effects on attitudes and behavior.”
The study, “Systematic Partisan Content Bias on TikTok During the 2024 U.S. Election,” was authored by Hazem Ibrahim, Daniel Jang Hyun-suk, Nouar Aldahour, Aaron R. Kaufman, Talal Rahwan, and Yasir Zaki.
