Social Media Algorithms May Show Users Content That Goes Against Their Beliefs, Study Finds

Social media algorithms may be promoting posts that conflict with users' personal beliefs and values, even as platforms claim to personalise content based on individual interests. A new study has found that content promoted by X's recommendation system often does not align with what users say they believe in.

The research, published in the Proceedings of the National Academy of Sciences (PNAS), suggests that social media algorithms may prioritise content that attracts attention and engagement rather than posts that reflect users' stated preferences.

The study was led by researchers Ziv Epstein of the Massachusetts Institute of Technology (MIT) and Michael S. Bernstein of Stanford University, along with their colleagues.

To examine how social media algorithms shape user experiences, the researchers studied 715 adults in the United States who actively used X, formerly known as Twitter.

Participants installed a customised browser extension that collected posts appearing in two different feeds: the 'Following' feed, which displays posts from accounts users choose to follow, and the 'For You' feed, which relies on an algorithm to recommend content.

The researchers then used an artificial intelligence model to analyse the posts and identify the values they reflected. They compared these findings with the personal values reported by participants.

The analysis revealed a clear difference between the two feeds. Posts appearing in users' Following feeds generally reflected their personal beliefs and values. In contrast, the algorithm-driven For You feed showed a tendency to promote content that did not correspond with those values.

Researchers found an overall negative relationship between users' stated values and the values represented in the content recommended by the algorithm.

The study also found that the algorithm appeared to favour posts related to order, tradition, rule-following and social security. Meanwhile, content associated with caring for others, dependability and environmental protection received comparatively less attention.

The researchers further examined whether political beliefs influenced these patterns. They found that users across the political spectrum were exposed to content that did not fully align with their stated values. However, the mismatch appeared to be more pronounced among Democratic users.

Why Do Algorithms Promote Content Users Disagree With?

The researchers believe users' own engagement patterns may help explain the phenomenon.

They analysed user activity over a 24-hour period, including likes, reposts and replies.

The findings showed that likes and reposts generally reflected users' personal values, indicating that people were more likely to endorse or share content they agreed with.

Replies, however, followed a different pattern. Users frequently responded to posts that did not align with their beliefs, often to challenge, criticise or argue against the content.

This behaviour may create an unintended consequence.

Social media algorithms commonly treat engagement as a signal of interest. As a result, when users repeatedly reply to posts they dislike, platforms may interpret those interactions as evidence that the content is relevant to them and recommend similar posts in the future.

Consequently, users may end up seeing more content that frustrates or upsets them simply because they engage with it to express disagreement.

According to the researchers, this pattern was particularly noticeable among Democratic users.

What Does This Mean For Social Media Users?

The findings raise broader questions about how social media platforms determine what appears in personalised recommendation feeds.

While many users assume that recommendation systems learn primarily from their interests and preferences, the study suggests that these systems may also respond to arguments, disagreements and other forms of negative engagement.

Researchers also found that globally popular content tended to align more closely with Republican values. This may help explain why right-leaning users were more likely to encounter content that matched their beliefs without engaging in confrontational exchanges.

However, the researchers cautioned against drawing firm conclusions about causation.

Because the study relied on observational data, the team analysed users' natural behaviour without experimentally altering the platform's algorithm. As a result, the research cannot conclusively determine whether replies to disagreeable content are the primary cause of the mismatch between users' values and the content they see.

The authors said further research involving controlled experiments and larger participant groups will be required to identify the exact mechanisms behind the phenomenon.

The study highlights a broader concern about social media platforms: algorithms designed to maximise engagement may not necessarily promote content that reflects users' genuine values and preferences.