A study published in the Proceedings of the National Academy of Sciences has found that X's recommendation algorithm favours "ragebait" content to boost engagement, and that this effect disproportionately affects users who identify as Democrats. The research, led by Stanford's Ziv Epstein, suggests the platform's feed algorithm learns from angry replies far more than from likes, creating a feedback loop that pushes ever more outrage-inducing posts to users, regardless of their stated values or beliefs.

Researchers tracked 715 American X users via a browser extension monitoring their For You and Following feeds, combining this with a "values inventory" based on the Schwartz Theory of Basic Values and self-reported political affiliations. They found the algorithm was more likely to amplify posts conflicting with users' own values, and that replies—though under seven percent of total interactions—carried disproportionate weight in shaping what content was subsequently shown. The reasons behind the stronger effect on Democrat-identifying users remain unclear, though researchers speculate it may relate to a greater volume of right-leaning content on the platform or a tendency among Democrats to engage more with posts they disagree with. X's former head of product, Nikita Bier, responded that the company had already reduced this effect by boosting friends' posts in the reply predictor last month; X itself has not formally commented.