What counts as a citation in AI search

If you're tracking AI citations, the first thing to get straight is that the three big engines don't agree on what a citation is. (For the bigger picture this fits inside, see our intro to Answer Engine Optimization.)

ChatGPT (web_search)

When ChatGPT browses, it returns a set of sources behind the answer. Some are inline-linked in the response. Others are listed but not linked from any specific sentence. We count both — a source the model read is a source the model considered, even if it didn't get a hyperlink.

Perplexity (Sonar)

Perplexity is the strict one. Every claim is footnoted to a numbered source. The footnote list is the citation set, and the order matters — the top sources tend to drive the framing of the answer.

Google AI Overviews

AI Overviews surface a panel of cited sources next to the generated summary. The panel is what we count. Google occasionally inlines a source mid-answer, but the panel is the canonical citation list.

Why the differences matter

If you only check ChatGPT, you'll think you're cited more than you are — ChatGPT's source lists are the longest. If you only check Perplexity, you'll think you're cited less — its lists are the shortest and most editorial. AI Overviews sits in the middle.

The honest measurement is to check all three, count citations consistently, and report on each engine separately. Averaging across engines hides the story. The shift this measurement enables — and why it matters now — is laid out in AI visibility vs. SEO.