Arguing with people on X may be teaching the platform to show users more of the posts they dislike.
A new study published Monday in the Proceedings of the National Academy of Sciences found that X’s "For You" algorithm amplified content that clashed with users’ stated values. The pattern was especially pronounced among Democrats.
"Social media feed algorithms rank content that is purported to be preferred by users, but the engagement behaviors that drive these algorithms are (at best) indirect proxies for users’ explicitly self-stated values," the researchers wrote.
Researchers tracked 715 active U.S. X users
According to the report, researchers from the Massachusetts Institute of Technology, the University of Michigan, Johns Hopkins University, Stanford University, and Yale University studied 715 active U.S. X users in September and October 2024. Participants installed a browser extension that collected their "For You" and Following feeds.
The study found that X showed Democrats more content that ran counter to their stated values, while the effect was much weaker for Republicans. The correlation was -0.229 for Democrats and -0.041 for Republicans, with more negative values indicating greater misalignment.
Across all users, amplified content scored -0.150, while posts from accounts users had chosen to follow generally aligned with their values.
"However, beyond this alignment between self-stated values and inventory, we observe participants’ values to be negatively correlated with the likelihood of content amplification," the researchers said. They added that the finding suggests "the algorithm is actually amplifying content that is misaligned with users’ explicit values."
Replication showed a similar partisan split
A replication produced a similar divide: -0.199 for Democrats and 0.0372 for Republicans.
Replies may explain part of the difference, according to the paper. Researchers said users who reply to people posting content that clashes with their values could inadvertently signal to X that they want more of it.
"Taken together, these findings offer a glimpse into the tensions between the values that people hold and those that provoke reactions, and how these value tensions can produce misaligned outcomes," the researchers wrote.
The study did not test whether the accounts were bots
The researchers did not examine whether the accounts behind posts that prompted value-misaligned replies were human or automated, leaving the role of bots unresolved.
Paper lands after X open-sourced its Grok-powered recommendation system
The study comes after X open-sourced its Grok-powered "For You" algorithm in January. The release showed a recommendation system that ranks posts by predicting user actions, including likes and replies.
Musk said at the time that the algorithm "needs massive improvements."
After the release, Musk posted: "We know the algorithm is dumb and needs massive improvements, but at least you can see us struggle to make it better in real-time and with transparency. No other social media companies do this."