X Algorithm Amplifies Divisive Posts as Arguments Fuel Recommendations

X’s recommendation system uses signals such as replies, likes and reposts to predict what will keep users engaged. New research suggests the platform may struggle to distinguish approval from outrage: when people argue with posts that conflict with their values, the interaction can be interpreted as interest. That dynamic matters because it may reward ragebait, amplify polarizing material and leave users with feeds increasingly shaped by content they dislike.

The study found that replying to objectionable posts can prompt X to recommend more content likely to provoke similar negative reactions, creating a cycle in which arguing produces further exposure. The findings indicate that engagement-based ranking can treat hostile or corrective responses as strong preference signals, even when users are expressing disagreement. The result highlights a central weakness in recommendation systems optimized for interaction rather than satisfaction or stated user intent.