A landmark scientific study has revealed that social media platforms possess the power to either inflame or reduce political polarization through algorithmic choices. Research on X demonstrated that barely perceptible adjustments to users’ feeds can generate as much political division in one week as naturally developed over three years, while also proving that similar adjustments can significantly decrease political animosity between opposing groups.
The research team conducted an innovative experiment involving more than 1,000 users during the 2024 presidential election. Using artificial intelligence to analyze posts in real-time, they modified what appeared in participants’ “for you” feeds. Some users received marginally more content featuring antidemocratic sentiments and partisan hostility, while others saw less of such material. The modifications were designed to remain imperceptible, and most participants indeed remained unaware their feeds had been altered, yet the psychological effects were substantial.
Martin Saveski, assistant professor at the University of Washington and lead researcher, emphasized that the algorithm’s power stems from its subtlety—barely noticeable changes resulted in significant shifts in users’ feelings toward political opponents. His colleague Tiziano Piccardi noted that the observed changes correspond to roughly three years of polarization based on historical trends in American political attitudes. The platform has faced scrutiny during the campaign for viral spread of manipulated and AI-generated content.
The measurement approach involved participants rating their feelings toward opposing political parties on a “feeling thermometer” scale ranging from 0 to 100 degrees. After one week, those exposed to more divisive content showed increased hostility of more than two degrees—matching the polarization increase that occurred over four decades from 1978 to 2020. Critically, the study also found that reducing divisive content decreased polarization by a similar amount, demonstrating that platforms could actively promote political harmony through algorithmic redesign.
These findings arrive at a crucial moment when democratic societies face unprecedented political division. Polling indicates that majorities believe people cannot agree even on basic facts, with many viewing current division levels as dangerous. While platforms have long been accused of amplifying divisive content to maximize engagement and advertising revenue, the research found that down-ranking such content resulted in only slight decreases in overall engagement volume. Moreover, users exposed to less divisive content actually engaged more meaningfully through likes and reposts. This suggests platforms could pursue social responsibility without completely sacrificing business viability, though it would require prioritizing societal well-being over maximizing every engagement metric. The researchers noted that platforms now have a practical method they could integrate to mitigate harmful personal and societal consequences.