X’s algorithm learns what you hate and shows you more of it, according to a new study just published in the Proceedings of the National Academy of Sciences (PNAS). The paper, titled Value misalignment of X’s feed algorithm is a reflection of value tensions in engagement, found that the site’s algorithm prioritized engagement above all else when it generated a user’s For You Page. It also showed that X serves more ragebait to people who say they are Democrats, although the exact reason for that is unclear.
“In 2026 that’s maybe not the most surprising headline ever,” Ziv Epstein, a postdoctoral researcher at Stanford University, and co-author of the paper, told 404 Media. “So we actually dug in a little deeper to figure out why this is actually happening, and it turns out that 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 study’s goal was to understand how a user’s self-professed values system might shape what they see on X. “We recruited a nationally representative sample of N = 715 Americans who are active users of X in September and October 2024, quota matched on ethnicity, gender and partisanship, to install a browser extension to collect their [For You Page] and Following feeds,” the study said.
Epstein said the study was observational and meant to get people asking questions about what they want to see on social media, how their feed is designed, and by whom. “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. “And so, we were very interested in trying to understand the particular effects of this particular algorithm, and so I think that has kind of important implications for civil society and just fighting some of the technofeudalistic tendencies of platforms to control these algorithms.”
For the study, researchers collected a “values inventory” of the volunteers using a research tool called the Schwartz Theory of Basic Values. The values inventory in the study is presented as a wheel with 19 points that corresponded to features like “tolerance,” “dominance,” “hedonism,” and “openness to change.” Users also reported their political alignments.
Then researchers watched how users engaged with posts on X and how those posts reflected their self-reported values. “We observe that the inventory of posts from followed accounts reflects users’ self-stated values — but that there is an overall negative correlation (misalignment) between users’ explicit values and the values in content that the algorithm is more likely to amplify,” the study said.
When a user on X sees a post that makes them mad — like a press release from a politician from a political party they don’t like — sometimes they’ll fight about the post in the replies. It doesn’t matter who you follow or what your stated values are, X reads replying as engagement and will send more of the infuriating posts the user’s way.
“When we look at commenting, the act of replying to posts, that's where we actually see some kind of meaningful misalignment between people’s values and the values of the content they’re replying to,” Epstein explained.
Most of the participants liked and reposted content on X and got served more of the same sort of content. Replying was rare, just 6.8% of the interactions according to the study, but had an outsized impact on the algorithm. “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 said. “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 and that seems to be particularly true of the Democratic users of our study.”
Though this happened across the political spectrum, the study found that ragebaiting occurred more for users that identified themselves as Democrats. “Democrat users confront the abundant value-misaligned content by replying to it, which the algorithm in turn preferentially learns from and continues to feed them,” the study said. “This highlights a core tension with how engagement-maximizing algorithms operate on social media: frictions between users’ stated preferences and their behaviors of reactive confrontation are exploited by engagement-maximizing algorithms to create runaway feedback loops of increasing value misalignment.”
Epstein said he’d need to do more research to find out why X seems to serve ragebait to Democrats more often than Republicans. It could be that there’s more rightwing content on X overall or it could be that Democrats tend to engage with posts they disagree with more often. “There might be some kind of differential effects on information diets there, or it might be something more psychological about how different you know partisan identities are triggering different kinds of actions and reactions, but ultimately I don't want to speculate too much,” he said.
I asked Epstein if he worried that prioritizing “values” in a social media algorithm might lead to more siloed user bases and more echo chambers. “I do think that if we go kind of down this path of thinking through and imagining value line social media feeds, we do have to be very aware of the potentials of value echo chambers, right?” he said. “Where people just, you know, they have the certain values that they have, and all the content they see is just aligned with those values. I think that is a very scary and dark reality.”
But he also said that values are intentional and that thinking about the kind of stuff you want to see on a social media site before you pull up your feed is a positive. “You're pumping the brakes, you're taking a breath, and you're thinking about what you actually really care about. In this world — and I don't have the data for this — but I would speculate that a lot of people actually do care about seeing a diverse set of content and engaging meaningfully across these lines, and you know maybe that isn't a particular value on my 19-dimensional wheel, but this is just a starting point of thinking about our intentions and thinking very deliberately about the kind of information we want to be exposed to, versus the the knee-jerk reaction that we're kind of learning in this very kind of short, shallow attention span, emotion and negative affect-driven model of these very myopic forms of engagement.”
X did not return 404 Media’s request for comment, but Nikita Bier — X’s former head of product — confirmed that the site’s algorithm had at one time been set to favor replies. “This is no longer true,” Bier said in a post on X. “The largest contributor of seeing ragebait was the reply predictor and we were aware that angry replies were causing people to see more of that content. So last month, we gave the reply predictor a 15x boost if it’s a friend’s post — and it reduced ragebait by [an] order of magnitude.”
Update 8/18/26 at 1:30PM: This story has been updated to include a comment from X’s former head of product
Connexion
