X Algorithm: Does Your For You Feed Clash With Your Values?

A study published in the journal PNAS followed 715 American adults who use X. The result: the "For You" feed, the one you scroll through, shows content that, on average, goes against their values. The mismatch is measured. Its cause isn't.

Key takeaways

• 715 American adults were tracked with a browser extension while they used X.
• The accounts they follow reflect their values. The "For You" feed is, overall, out of step with them.
• The feed pushes conformity, order and tradition more, and kindness or nature less.
• The authors don't prove that heated replies cause this mismatch.

Who was tracked, and how?

Participants installed an extension that captured posts from two feeds: the accounts they follow, unranked, and the "For You" feed, picked by the algorithm. An AI model identified the values expressed in each text, based on Schwartz's theory of basic human values, which lists 19 of them.

A 24-hour follow-up then recorded likes, reposts and replies. The study only covers the United States. It's observational: the researchers didn't change the algorithm to see what would happen.

The values the feed pushes, and the ones it sidelines

The "For You" feed slightly favors individualistic values over collectivist ones. It amplifies conservation and order: following the rules, keeping society safe, tradition. It dials down self-transcendence values, like kindness, dependability or protecting nature.

The mismatch affects both political camps, but it's stronger among Democrats. Content that is popular overall sits closer to Republican values.

Likes agree, replies don't

Likes and reposts generally line up with the user's values. Replies are less aligned. Among Democrats, they are on average slightly at odds with their values.

The authors' hypothesis comes down to a loop. We mostly reply to what annoys us, that reaction feeds the algorithm, which then shows more content of that kind. The authors present this loop as a possible explanation. They haven't tested it on the algorithm itself.

A proven mismatch, an unproven cause

The claim that "X's algorithm pushes content against your values" is partly confirmed. The mismatch shows up in the data, on the "For You" feed, for both political camps.

But the study doesn't say the algorithm is the cause. Without a controlled change, you can't separate what comes from the algorithm from what comes from the users themselves. The authors call for controlled experiments and larger samples.

And what about your feed?

The study doesn't offer a practical fix. It does give you a simple thing to watch for: if you notice you mostly reply to posts that bug you, that's exactly the mechanism the authors suspect.

On the link between screens and attention, dopamine detox, real or toc? sums up what science knows, and what it still doesn't.