Reddit is valuable because people tell you what actually happened to them. Its AI search may be sanding some of that away.
A new study from researchers at the University of Illinois Urbana-Champaign audited Reddit’s AI search using 10,000 queries run three times, producing 30,000 answers that researchers traced against 14.68 million comments.
Search Engine Journal spotted the study first, and the headline finding is already interesting: comments that were highly rated, prominent in a thread and written more formally were disproportionately likely to make it into an AI-generated answer.
But the more interesting part is what happens to Reddit’s human voice once the AI starts summarizing it.
Reddit turns experience into advice
The researchers found that comments containing stronger markers of personal experience were less likely to be selected in the first place.
Then the AI flattened much of what remained.
First-person singular language, words such as “I” and “my”, accounted for 3.3% of tokens in the source comments. In Reddit’s generated answers, that fell to just 0.06%, a 98% reduction.
The researchers describe this as a two-stage effect: personal experiences are less likely to be selected, then the experiences that do survive are rewritten into a more generalized voice.
That distinction matters on Reddit.
“This happened to me” is not the same kind of evidence as “this is what you should do.” Yet an AI summary can make both sound like authoritative advice.
Popular Reddit comments get another popularity boost
Reddit’s existing ranking signals also appear to carry into the AI layer.
A one-standard-deviation increase in a comment’s score percentile within a thread was associated with 2.88 times the odds of being selected. Top-level comments were also nearly twice as likely to appear.
In other words, comments Reddit users have already pushed toward the top get another chance to dominate when Reddit’s AI builds its answer.
That does not prove Reddit’s AI deliberately rewards popularity. The study is observational, and the researchers explicitly warn against interpreting these relationships as causal.
Small communities can get drowned out too
The effect is not limited to individual comments.
Reddit AI frequently builds an answer from multiple communities. For queries originating in the smaller communities studied, roughly two-thirds of retrieved posts came from larger communities.
The researchers also found that already-visible questions were easier for the system to recover.
So a question asked in a niche community does not necessarily produce an AI answer dominated by that community’s expertise. Reddit may pull the discussion back toward larger parts of the platform instead.
There is one more warning buried deeper in the paper.
Researchers separately examined 2,000 claims generated by Reddit Answers using two verification methods. 8.7% were classified as unsupported by both. The two verifiers disagreed substantially, so the authors say the true rate cannot be established from this experiment, but they consider the overlap evidence that some generated claims cannot easily be traced back to the sources Reddit presents.
Reddit without the Reddit?
This is the awkward part for Reddit’s AI strategy.
Reddit originally pitched Reddit Answers around quickly finding the “human perspectives” buried across its communities. Yet this study suggests the AI layer can favor comments that already look authoritative, reach into larger communities for answers and turn first-person experiences into generalized advice.
None of that makes Reddit AI search useless.
But if you’re using Reddit precisely because you want messy, conflicting accounts from actual people, the AI summary may not be a neutral shortcut to the discussion.
Sometimes the mess is the useful bit.
Start the conversation by posting the first comment