New research has provided definitive evidence of social media algorithms’ capacity to rapidly reshape political attitudes. Scientists found that subtle adjustments to X users’ content feeds generated levels of political polarization in one week that historically took three years to develop, demonstrating platforms’ extraordinary influence over how people view political opponents and engage with democratic discourse.
The experiment was conducted during the 2024 presidential election campaign and involved more than 1,000 X users. Researchers employed artificial intelligence to analyze posts in real-time, classifying them based on characteristics likely to increase division. They then manipulated participants’ “for you” feeds to show either more or fewer divisive posts. The targeted content included material expressing antidemocratic attitudes, partisan violence, opposition to bipartisan consensus, and biased evaluations of politicized facts. Most participants remained unaware their feeds had been altered.
The election period witnessed numerous viral incidents on the platform, including widespread dissemination of manipulated images and AI-generated propaganda. Since the company’s acquisition and rebranding, X has employed algorithmic curation that prioritizes engagement-maximizing content over simply showing posts from accounts users actively follow. This shift has generated ongoing debate about the platform’s role in political culture and democratic institutions.
After one week of exposure to modified feeds, researchers measured participants’ attitudes using a “feeling thermometer” approach. Participants rated their warmth or coldness, favorability or unfavorability toward political opponents on a scale from 0 to 100 degrees. Those who saw more divisive content exhibited increased negative feelings of more than two degrees—precisely the amount of polarization that developed across American society during the four decades leading to 2020. The study also found that repeated exposure to antidemocratic and partisan content significantly boosted feelings of sadness and anger.
The research offers both concerning revelations and hopeful possibilities. While it confirms that platforms can rapidly intensify political polarization, it also demonstrates they could just as easily reduce it through algorithmic design choices. Users who saw fewer divisive posts exhibited decreased polarization by a similar magnitude. Although platforms have been suspected of amplifying divisive content to boost engagement and advertising revenue, the study found that down-ranking such content produced only modest reductions in time spent and posts viewed, while actually increasing meaningful engagement through likes and reposts. This suggests platforms could integrate methods to mitigate harmful consequences while maintaining viable business models, though success would require prioritizing societal well-being alongside profit maximization.
