Why misinformation spreads faster on X: the algorithm adds all engagement into a single score

That misinformation on social media travels faster than verified information has been known for years. What was missing was the mechanical reason. A study by Pan Li and Shuang Gao, published as a preprint on 24 September, looks for it inside X's own recommendation algorithm, which they examine component by component.

The mechanism: all engagement goes into the same sum

The algorithm predicts what each user will do with a post (like it, repost it, reply or quote it) and combines those predictions into a single score, a weighted sum. The authors call this engagement fungibility: for the score, one reaction can be swapped for another.

The consequence is that a post can be amplified through instant reactions even when hardly anyone stops to reply. And misinformation attracts that kind of reaction above all.

What they tested

Using X data on the 2024 US elections, they found that adjusting the weight of each metric changes very little. What does work, according to the study, is a prior condition: content is not amplified until the system predicts a minimum of thoughtful engagement. Across 46 robustness checks, that gate reduced exposure to low-credibility content without reducing exposure or engagement for mainstream content.

Why it matters beyond X

Because almost everyone measures engagement the way that algorithm does: by adding it up. And two posts with the same total can be very different things. One prompted a thousand passing likes. The other, two hundred replies from people who stopped to write.

How we read it at Trendia

When we analyse a trend we look at what kind of response it gets, not only how much. It helps to separate three things:

• Quick reaction: like, view, share without comment.
• Thoughtful response: a comment with an argument, a quote, a reply video.
• Amplification: how much new exposure follows.

That is why comments are part of the analysis and not an appendix. In the project with the UOC on mental health on TikTok, the corpus included more than 2,200 videos and more than 67,000 comments.

One caution: the study is about X and a specific electoral context. The TikTok and Instagram algorithms are not public, so it cannot be assumed they work the same way. What does carry over is the question: is this engagement peak interest, or is it reflex?

You can see how we work on a real project.

Source: Why Does Misinformation Propagate Faster? An Algorithmic Perspective on X, arXiv, 24 September 2026.