Algorithmic Feed Changes Show Polarisation Isn’t Inevitable

A new study led by researchers at Stanford demonstrates that social media polarisation can be eased through modest adjustments to content ranking rather than content removal or heavy moderation. The team developed a browser-based tool that reorders users’ feeds on X to push down posts containing extreme partisan hostility, calls for political violence or antidemocratic rhetoric. When applied in a field experiment during the 2024 US election, this intervention softened negative attitudes toward opposing political groups across the partisan spectrum.

The experiment involved 1,256 individuals who consented to have their feed content re-ranked over a 10-day period. Some participants had exposure to aggressive and antidemocratic posts reduced, while others experienced an increased exposure. Participants in the reduced-hostility group exhibited a shift toward warmer feelings for those supporting the other party, while those exposed to more hostility became colder. The change, though modest in absolute terms, is equivalent to about three years of polarisation as historically measured between 1978 and 2020.

The research underscores that platform algorithms play a decisive role in shaping political discourse. By using a large-language model to flag and re-score content in real time, the tool identifies posts that breach democratic norms or express intense partisan animosity — without removing any content or requiring cooperation from X itself. All posts remain available, but those judged toxic appear later in the feed hierarchy, reducing their visibility.

The findings challenge the notion that political toxicity and division are an unavoidable byproduct of social media. Instead, they suggest that design decisions — particularly around algorithmic ranking — influence how divisive content spreads and how users perceive opposing viewpoints. Crucially, the intervention did not significantly alter standard engagement metrics like likes or reposts, indicating that it may be possible to curb polarisation without sacrificing platform activity or user engagement.

Authors of the study believe this opens the door to giving users and researchers more control over the algorithms shaping their experience — a shift away from opaque, engagement-driven systems toward more socially purposeful curation. They envisage a future in which individualized feed algorithms prioritise civic health and cross-party empathy rather than outrage and conflict.