PNAS Study Finds X Algorithm Heavily Amplifies Outrage and Targets Democratic Users

In conducting a study on platform algorithms, scholars monitored 715 American X users who installed a custom browser extension to capture content served on their “For You” and “Following” feeds. Participants completed a detailed values inventory based on the Schwartz Theory of Basic Values—a psychological framework developed by researcher Shalom H. Schwartz that categorizes human motivations across 19 distinct universal values, including qualities such as tolerance and dominance. Volunteers in the cohort also disclosed their personal political affiliations.

Researchers subsequently analyzed how participants interacted with individual posts and evaluated how those posts corresponded to their self-reported personal values. The empirical data revealed that X’s recommendation engine was “more likely to amplify” posts that directly conflicted with a user’s core values. Consequently, participants were rapidly exposed to provocative material that sparked frustration and anger.

These findings form the core of a peer-reviewed study published in the Proceedings of the National Academy of Sciences (PNAS), which confirmed long-standing concerns that X’s recommendation algorithm systematically elevates ragebait to maximize user engagement. Furthermore, the empirical investigation demonstrated that outrage-inducing content was delivered at higher volumes to users who identify as Democrats.

The heightened exposure to ragebait among self-identified Democratic users stood out as a key empirical pattern. While the researchers noted they do not yet possess a definitive explanation for this disparity, they hypothesized that it may stem from a higher volume of right-wing content circulating on X following recent platform shifts, or from a behavioral tendency among Democrats to actively respond to posts expressing views they oppose.

When an inflammatory post provokes anger in a user, that individual becomes significantly more likely to interact with it, triggering the algorithm to funnel increasingly outlandish content into their feed. Lead author Ziv Epstein—a computer scientist and postdoctoral researcher at Stanford University—pointed out that replying to an unfounded or offensive post carries a far heavier algorithmic weight than simply liking a post, even though replies comprise less than seven percent of all user interactions on the platform.

“Replying is only a fraction of engagement, but there does seem to be some evidence that these algorithms are prioritizing and learning more from this kind of rarer form of engagement,” Epstein stated. “So it’s this feedback loop of outrage baiting. The algorithm learns that you get outraged and then continues to serve more content in that direction.”

Speaking to tech news outlet 404 Media, Epstein acknowledged that while the core conclusion is “not the most surprising headline ever,” the research team sought to pinpoint the underlying mechanics driving the platform’s behavior. “X’s feed algorithm, like a lot of these social media algorithms, is optimized for engagement [but] it turns out that not all types of engagement are considered equally,” the Stanford investigator explained.

Beyond documenting algorithmic bias, Epstein noted that the research aims to stimulate broader public debate regarding platform architecture and algorithmic control. “There are these social media algorithms that have enormous amounts of power in our lives, they shape the information that we consume and we have very little transparency into how they operate and what their implications are,” he said. “I think that has kind of important implications for civil society and just fighting some of the techno-feudalistic tendencies of platforms to control these algorithms.”

These scientific findings capture an everyday reality familiar to anyone who has monitored X over recent years. Standard user feeds frequently present an unrelenting stream of right-leaning provocative accounts expressing outrage over controversial cultural topics, such as retail stores at the Mall of America offering hijabs for sale.

X has not issued an official corporate response to the study. However, tech entrepreneur and former X product executive Nikita Bier stated that platform engineers modified the recommendation algorithm after the research data was collected, claiming ragebait content was subsequently curtailed “by order [sic] of magnitude.” While ongoing user experiences indicate outrage-centric content remains prominent, Bier’s comments indicate platform management recognizes the dynamic and has taken steps to address recommendation parameters.