Overview
- In a Science study, 1,256 consented X users were randomly assigned feeds with increased, decreased, or unchanged levels of posts expressing antidemocratic attitudes and partisan animosity.
- Downranking hostile political content made participants feel more than 2 points warmer toward the opposing party on a 100-point scale, while upranking produced comparable colder shifts.
- The effects were bipartisan and accompanied by in-feed changes in emotions such as anger and sadness, though durability beyond the intervention remains unknown.
- The intervention used an LLM-based classifier to score content and a lightweight browser extension to reorder posts in real time without platform cooperation.
- The research team released the tool’s code to enable replication and new studies, highlighting a method for user- or researcher-controlled feed experiments across platforms.