A landmark study has documented the remarkable speed at which social media algorithms can generate political polarization. Research on X demonstrated that subtle adjustments to users’ content feeds created as much political division in one week as naturally developed over three years historically, revealing the platform’s extraordinary capacity to shape political attitudes and democratic discourse.
The experiment utilized cutting-edge artificial intelligence to analyze and manipulate content in real-time. Scientists examined posts in X’s “for you” feed and adjusted what appeared to more than 1,000 participants during the 2024 presidential election. Some users saw marginally more posts containing antidemocratic sentiments and partisan hostility, while others saw fewer such posts. The changes were designed to remain barely perceptible, and indeed most participants didn’t notice their feeds had been altered, yet the psychological effects were substantial and measurable.
The platform has faced increasing scrutiny for its role in political discourse since its acquisition and rebranding. The introduction of the algorithmic “for you” feed, which prioritizes engagement-maximizing content over posts from followed accounts, has coincided with viral spread of manipulated images and inflammatory political content. The campaign period saw numerous examples of AI-generated propaganda and fake images that garnered tens of millions of views.
Researchers measured changes in political attitudes using a rigorous methodology. After one week of exposure to modified feeds, participants rated their feelings toward political opponents on a “feeling thermometer” ranging from 0 to 100 degrees. Those exposed to more divisive content exhibited increased negative feelings of more than two degrees on this scale—precisely matching the polarization increase that occurred across four decades of American history from 1978 to 2020. The study also found that reducing divisive content decreased polarization by a similar amount.
These findings have profound implications for addressing political division in democratic societies. Survey research shows that vast majorities believe political opponents cannot agree on basic facts, and many view current divisions as dangerous to society. The study proves that platforms possess the technical capability to reduce this polarization through algorithmic design choices. While social media companies have been suspected of amplifying divisive content to boost engagement and advertising revenue, the research found that down-ranking such content produced only slight decreases in time spent and posts viewed, while actually increasing meaningful engagement through likes and reposts. This suggests platforms could integrate polarization-mitigation methods without completely sacrificing their business interests.