In one sentence: count only the sources marked in the answer body, rank your domain among all cited sources rather than reporting a share, and read the source mix as a map of where the engine goes looking — which is usually more actionable than your own position in it.
The distinction that decides whether the number is real
An engine assembles a retrieval candidate pool — often a dozen or more documents — and then writes an answer citing a subset. Counting the pool instead of the in-answer markers overstates citation share by a multiple. On our own runs it was two to four times before we fixed it.
A retrieval pool is not a citation. If a tool cannot tell you which of the two it counted, assume the larger one.
Report rank, not share
"5% citation share" is unstable: it moves when the engine cites more or fewer sources overall, which has nothing to do with you. "Fifth most-cited source out of 109" is an ordinal over a real population — it moves when your position actually changes, and it is immediately interpretable by someone who has never seen the methodology.
The source mix is the real finding
Aggregate every cited URL across your question set and group by domain and by type. This tells you where the engine looks for this category, which is a content distribution plan you did not have to guess at:
| Source type | What a high share implies |
|---|---|
| Community threads | Presence in discussions matters more than your own pages |
| Reference and encyclopaedic sites | Entity clarity and third-party descriptions carry the weight |
| Vendor documentation | Your docs are a first-class surface, not a support afterthought |
| Review and directory sites | Listing completeness is doing work your marketing site cannot |
The mix differs by engine and by question type, so read it per engine. An engine that leans on community threads will not reward the same work as one that leans on documentation.
Your own domain: which pages, not how many
When your domain is cited, record which URL. Two patterns are worth acting on. If one page is carrying every citation, the rest of the site is not extractable and that page is a single point of failure. If citations land on pages you did not intend as entry points — an old blog post, a glossary entry — that is the engine telling you which of your content is actually usable, and it is usually not the page you optimised.
What citation data cannot tell you
Whether you were recommended. A citation means your page was used as reference material; the answer may still have suggested a competitor. Track the two separately or you will read a rise in citations as a win when it is sometimes the opposite — your content informing an answer that sends the buyer elsewhere.