What's this about?
A separate audit from the manufactured-sources report, published the same day: roughly a third of Perplexity's citations do not contain the number they are cited for.
Why is a wrong citation worse than a wrong answer?
Because the citation is the trust mechanism. A model that says something wrong invites scepticism; a model that says something wrong *and shows a source* converts scepticism into confidence, and almost nobody clicks through. The failure is worse than hallucination because it launders the error through an apparatus designed to prevent exactly that — and it is undetectable to the reader without doing the work the citation was supposed to save.
What's the community saying?
HN 124 points, 45 comments
- A first-hand report matching the finding: a citation previewed as a reliable source turned out to reference a Reddit thread and other weaker sources.
- The most useful exchange is about whether this is fixable: correct externally-validated citations are one of the first things you build when you give a model document access, so a large public service missing this is a credibility problem.
- The proposed fix is deterministic verification — force the system to check that the cited page contains the claim, with a tool rather than a model.
- A genuinely fair defence: Perplexity provides more references than competitors, which makes its errors easier to evaluate. A similar error rate without visible sources would be worse.