X algorithm boosts ragebait — and Democrats are hit hardest, study finds

A study published in the Proceedings of the National Academy of Sciences (PNAS) has confirmed what many users have suspected since Elon Musk took over the platform: X's algorithm prioritizes ragebait content to drive engagement, and this content disproportionately affects users who identify as Democrats.

The research involved 715 American X users who installed a browser extension that collected data from their For You and Following feeds. Participants were asked to complete a "values inventory" based on the Schwartz Theory of Basic Values, which breaks down belief systems into 19 points, including qualities like "tolerance" and "dominance." Volunteers also reported their political affiliations. Researchers then tracked how these users engaged with posts and how those posts reflected their stated values.

One of the paper's authors, Ziv Epstein from Stanford University, acknowledged that the findings are "not the most surprising headline ever" but emphasized the importance of understanding the underlying mechanisms. "X's feed algorithm, like a lot of these social media algorithms, is optimized for engagement it turns out that not all types of engagement are considered equally," Epstein told 404 Media.

The study found that the algorithm systematically promotes provocative content that triggers strong emotional reactions, particularly anger. This type of content generates more likes, reposts, and comments — key metrics for the algorithm — and is served more frequently to Democratic users. The effect is likely amplified during politically charged periods, as the platform's design favors divisive material over neutral or constructive posts.

The findings align with widespread observations of X's feed since Musk's acquisition, where users often report seeing a stream of right-leaning troll accounts and inflammatory content. The study provides empirical evidence that the platform's engagement-driven model exacerbates political polarization, especially among those who are already more likely to be targeted by such content. The authors argue that this dynamic is not accidental but a direct result of the algorithm's optimization for metrics that reward outrage over thoughtful discourse.