All AI search engines massively distort quotes.

According to a Columbia Journalism Review study, more than 60% of AI search engine answers contain citation errors: headlines, dates, publisher names, and source URLs are distorted. Even paid versions (Perplexity Pro, Grok 3) confidently provide incorrect information instead of admitting gaps in the data.

The only things I can nitpick are the small sample size and that the search was run once. But that’s unlikely to change the result much. LLMs don’t handle single pieces of data precisely enough. You can see it even in programming. And how fixable that is is unclear.