Social media algorithms can alter political views, browser extension study shows (www.euronews.com)

Researchers from Stanford, the University of Washington and Northeastern published a controlled experiment in Science showing that an AI-powered browser extension can measurably change users’ political attitudes by reordering content in their X (formerly Twitter) feed. The extension uses AI to scan posts for anti-democratic or extremely negative partisan themes (e.g., calls for violence or jailing opponents) and demotes them within seconds. Over a 10-day trial with more than 1,200 consenting participants ahead of the 2024 US presidential election, participants exposed to feeds with down-ranked hostile content reported an average 2-point improvement on a 1–100 scale toward the opposing party—an effect the authors equate to roughly three years’ worth of change in US affective polarization. Emotional responses (less anger and sadness) also declined while the extension was active. The work is significant because it demonstrates a method for independent researchers to study and manipulate algorithmic ranking effects without platform cooperation, and suggests straightforward interventions platforms could deploy to reduce partisan animosity. Effects were bipartisan and emerged quickly, but the study is limited to browser users (not app audiences) and measured only short-term changes; emotional benefits dissipated after the experiment ended and long-term impacts remain unknown. The paper argues for reranking as a practical lever to promote healthier civic discourse, while highlighting the need for further work on persistence and broader platform integration.