X's Algorithm Loves "Replies" to Angry Posts: Stanford Study Says Democratic-Leaning Users Get Provoked More Often

Replies make up less than 7% of all engagement on X, yet they're the signal the platform's algorithm values most. A study published in the Proceedings of the National Academy of Sciences (PNAS) identifies this as the key mechanism behind why "ragebait" content keeps spreading on the platform—and it shows up more often for users who identify as Democratic supporters.

The research team recruited 715 US-based X users and had them install a browser extension to collect content from their "For You" and "Following" feeds. Before the study, participants filled out a "values inventory" based on the Schwartz Theory of Basic Values, which breaks down personal beliefs into 19 traits like "tolerance" and "dominance," and also self-reported their political leanings. Researchers then compared what these users actually engaged with against the values they'd reported about themselves.

The results showed the algorithm "tends to amplify" content that conflicts with users' own values—in other words, people quickly end up seeing things that make them angry. Ziv Epstein, a Stanford researcher and co-author of the paper, told 404 Media: "X's feed algorithm, like a lot of social media algorithms, is optimized to boost engagement, but the result is that not all engagement is treated equally." He explained that replies are weighted far more heavily by the algorithm than likes, even though replies account for less than 7% of engagement on the platform. "This creates a ragebait feedback loop—the algorithm learns you can be provoked, and it keeps feeding you more content in that direction."

The study also found that this pattern of being provoked shows up more often among users who identify as Democratic supporters, but the research team doesn't have a definitive answer yet—only two possible explanations: either right-leaning content is simply more prevalent on X, or Democratic supporters tend to engage with posts they disagree with. Both remain speculative, and the paper doesn't confirm either one.

Epstein said one goal of the study was to get more people talking about how social platforms actually operate: "These social media algorithms hold enormous power over our lives—they shape the information we receive—but we have almost no transparency into how they work or what effects they have."

X has not officially responded to the study, but former head of product Nikita Bier posted on his own X account that the reply-prediction mechanism is the biggest source of users seeing so much angry content, saying, "Last month we increased the weight of friends' posts in the reply predictor by 15x," and claimed the change has already led to a "significant reduction" in such content. This adjustment has only appeared in his personal posts so far, with no official platform announcement to back it up.