Particle.news

Study Finds X Feed Amplifies Posts That Clash With Users' Values

Researchers say rare reply interactions teach the recommendation system to boost content that provokes anger, a pattern that could distort users’ information diets.

Overview

  • A peer-reviewed paper published on August 18, 2026 analyzed data from 715 U.S. X users and found that the platform’s For You algorithm amplified content that was negatively correlated with users’ self-reported values.
  • The measured misalignment was stronger for self-identified Democrats, with a reported correlation of −0.229 for Democrats versus −0.041 for Republicans, and a replication showing a similar partisan split.
  • Researchers identify replying as a key driver: replies made up about 6.8% of recorded interactions but had an outsized effect on what the algorithm learned and recommended.
  • The team used the 19-point Schwartz values inventory and passive feed collection via a browser extension in Sept–Oct 2024, and they stress the findings are observational and do not prove why the partisan gap exists.
  • Coverage across outlets highlighted concerns about outrage-driven feedback loops and called for more platform transparency and follow-up work, while X did not offer substantive comment in the reports.