Overview
- The peer-reviewed paper published Tuesday in PNAS analyzed feeds from a quota‑matched sample of 715 active U.S. X users whose data were collected in September–October 2024.
- Researchers measured users’ self‑stated values with the Schwartz inventory and found that posts from accounts users follow generally matched those values while algorithm‑amplified posts showed an aggregate misalignment of about −0.150.
- The study reported a clear partisan gap in misalignment, with correlations of roughly −0.229 for self‑identified Democrats and −0.041 for Republicans, and a replication that produced a similar split.
- Authors identified a mechanism where replies—just 6.8% of interactions—are heavily weighted by the Grok‑style recommender, causing ‘ragebait’ feedback loops, but they stress the work is observational and did not test whether bots drove the effect.
- The paper follows X’s January open‑sourcing of its Grok recommender and Elon Musk’s public comment that the algorithm needs major fixes, and it has renewed calls for more research and clearer platform transparency while X has not publicly responded.