Not impressed with a political post on X? Elon Musk’s social media platform may show you more of those, says new study

A new study has revealed how X's algorithm handles user engagement through ‘hate management’. Published recently in the Proceedings of the National Academy of Sciences, the research uncovered concerning patterns. The paper is titled "Value misalignment of X's feed algorithm”.

Researchers have found that the algorithm prioritises engagement above almost everything else. This constantly shapes what users see on their personalised feeds. The study also revealed that X serves more provocative content to certain users.

Here is an interpretation of the results of this research in the Indian context. Suppose you don’t like the CJP movement. You go and comment under a post by Abhijeet Dipke.

Your comment clearly shows that you’re not impressed by them. X will show you more posts from the CJP founder or from similar people.

In the US, Democrats reportedly receive more "ragebait" than other political groups. However, researchers remain uncertain about the exact reason behind this pattern.

Ziv Epstein, a postdoctoral researcher at Stanford, co-authored this study. He explained to 404 Media that the team wanted a deeper understanding beyond surface-level observations.

According to him, engagement types aren't treated equally by the algorithm. Some interactions carry far more weight than others within X's system.

The research involved 715 active American users of X, owned by Elon Musk. Users took part during September and October 2024. Researchers matched participants based on ethnicity, gender and political affiliation.

Volunteers installed a browser extension allowing researchers to collect their feeds. This included both their "For You" and Following pages.

Epstein emphasised that this study remained purely observational. He hoped that it would encourage broader questions about social media design.

According to him, these algorithms hold immense power over daily life. They currently shape information consumption with very little public transparency.

To measure values, researchers used a tool called Schwartz Theory. This presented values as a wheel featuring 19 distinct dimensions. These included concepts like tolerance, dominance, hedonism and openness.

Participants also disclosed their political leanings for comparison purposes. Researchers then tracked how users engaged with various posts.

The study found a clear misalignment between stated values and content shown. Users followed accounts that generally reflected their personal values.

However, amplified content often contradicts these same stated values. This misalignment became particularly evident through user replies.

When users encounter posts that provoke anger, they sometimes respond with anger. X interprets these replies as valuable engagement, regardless of tone.

“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,” Epstein said.

Replies, however, remained relatively rare. It represented just 6.8% of interactions overall.

Despite their rarity, replies carried disproportionate influence over algorithmic behaviour. Epstein described this as a feedback loop that consistently encourages outrage.

Most users simply liked or reposted content without engaging further. This behaviour generated more similar content without triggering outrage loops.

The pattern appeared across the political spectrum, though unevenly distributed. Democrats specifically confronted misaligned content through replies more frequently.

This confrontation, researchers noted, exploited the algorithm's engagement-maximising design. It created runaway feedback loops of increasing value misalignment.

More research needed

Epstein has said that he needs further research to explain this partisan difference. He suggests various possible explanations without confirming any specific cause.

“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 told 404 Media.

When asked about echo chambers, Epstein expressed genuine concern about future risks. He warned that value-based feeds could create isolated content bubbles.

He described this potential outcome as genuinely troubling for society. However, he also saw value in intentional content consumption practices.

He encouraged users to reflect deliberately on desired content beforehand. This means pausing before automatically engaging with provocative posts.