TL;DR
A randomized field experiment on X found that subtly increasing the presence of antidemocratic and partisan posts in users' "for you" feeds shifted affective polarisation in a week by an amount comparable to roughly three years of historical US trends. Down-ranking the same kinds of content produced a similar reduction in polarisation, though it modestly lowered some measures of engagement.
What happened
Researchers from the University of Washington, Stanford, Johns Hopkins and Northeastern ran a week-long experiment on more than 1,000 X users during the 2024 US presidential campaign. Using an AI system to classify posts appearing in X's "for you" feed in real time, the team increased the exposure of one cohort to posts expressing antidemocratic views and partisan animosity and reduced such posts for another cohort. After seven days participants reported changes on a standard "feeling thermometer" measuring warmth toward political opponents; the shift in affective polarisation exceeded two points on a 0–100 scale, a change the authors say matches about three years of US polarisation observed between 1978 and 2020. Most participants did not notice the subtle feed alterations. The study also found that lowering divisive content cut polarisation by a similar margin, though it led to a small drop in time spent on the platform and posts viewed while increasing likes and reposts.
Why it matters
- Algorithms that curate feeds can quickly alter users' political attitudes at scale.
- Platforms have the technical capacity both to amplify and to reduce political animosity through ranking choices.
- Efforts to down-rank divisive content may reduce short-term engagement, creating a trade-off for ad-driven business models.
- Rapid shifts in affective polarisation have implications for public discourse and democratic stability.
Key facts
- Study published in the journal Science by teams at UW, Stanford, Johns Hopkins and Northeastern.
- More than 1,000 X users participated during the 2024 US presidential election period.
- Researchers used AI to identify and adjust the presence of posts with antidemocratic and partisan animosity in real time within the "for you" feed.
- A one-week exposure produced a reported change of over two degrees on a 0–100 feeling thermometer measuring affective polarisation.
- The measured effect size corresponds, according to the authors, to roughly three years of US polarisation trends from 1978–2020.
- Reducing divisive posts produced a comparable decrease in polarisation.
- Lowering divisive content slightly reduced time on platform and number of posts viewed, while increasing likes and reposts.
- Most participants did not detect that their feeds had been modified.
- The study cites examples of high-reach divisive posts on X during the campaign, including manipulated images that went viral.
- Elon Musk rebranded Twitter as X after buying it in 2022 and the platform uses a "for you" feed that elevates engagement-optimised content.
What to watch next
- Whether X updates its ranking algorithms or moderation policies in response to this research: not confirmed in the source.
- Regulatory or policy moves that address algorithmic amplification of political content: not confirmed in the source.
- Follow-up or replication studies that track longer-term effects beyond a single week: not confirmed in the source.
Quick glossary
- Affective polarisation: Emotional dislike or hostility between supporters of different political parties, often measured by warmth/feeling scales.
- Algorithm: A set of rules or calculations used by platforms to decide which content to show users.
- "For you" feed: A social platform feature that surfaces content selected by an algorithm rather than only by accounts a user follows.
- Down-rank / up-rank: To reduce or increase the visibility of specific posts in a user's feed via algorithmic weighting.
- Engagement: Measures of user interaction on a platform, such as time spent, likes, reposts and views.
Reader FAQ
Did the study show small feed changes can shift political views quickly?
Yes. The research found a one-week tweak produced an affective polarisation change roughly equivalent to three years of historical US change.
Who ran the experiment and where was it published?
Teams from the University of Washington, Stanford, Johns Hopkins and Northeastern authored the study, which appears in Science.
Did participants notice their feeds were altered?
Most of the more than 1,000 participants did not detect the subtle modifications to their feeds.
Did X respond to the findings?
X was approached for comment, but a platform response is not confirmed in the source.
View image in fullscreen Musk introduced the ‘for you’ feed in X, which uprates content calculated to maximise engagement. Photograph: Reuters X This article is more than 1 month old…
Sources
- Partisan X posts boost political polarisation among users, research finds
- Post-January 6 deplatforming shows long-term effects on …
- Influencers and Multipliers reinforce political Polarization
- Some social media platforms become more polarized …
Related posts
- Meet the new tech laws of 2026: state AI, privacy, repair and limits
- MHC: Manifold-Constrained Hyper-Connections for Scalable Residual Models
- AI Labor Is Boring, but AI Lust Turned Erotic Chatbots into Big Business