Citation rate: the share of answers that credit you

Assistants answer in front of the click, and an answer credits three or four sources. If yours is not among them you were not considered, and nothing in your analytics records the loss: a question answered without you produces no impression to explain.

That makes citation rate one of the few readings of demand your own reporting never sees.

• It moves independently of traffic. Sessions can fall in a week where you are cited more often, and rise while an assistant quietly settles on a competitor as its default source.
• It is earned per question, not per site. A supplier can be the cited authority on installation questions and invisible on pricing questions in the same month, which tells you precisely what to publish next.
• Incumbency is sticky, not permanent. Assistants retrieve from a search index and then tend to favour pages that state an answer plainly, so a source that gets picked often keeps getting picked — until a clearer or fresher page takes the slot.

A rank tracker sees none of this. Answer engine optimisation is the work; the citation rate is the scoreboard.

Every site the engine publishes to names its own list of the questions it wants to be the answer to. Those questions are per-site settings, because what matters to a CCTV wholesaler shares nothing with what matters to a Persian developer-tools site. A nightly job puts the list to GPT and Perplexity, records whether the site was named or linked, and files the result beside the same period's Search Console clicks and impressions.

Two caveats we hold to.

First, a probe measures page-level citation only if the assistant actually searches the web during the run. A model answering from memory is reporting brand recall, which is a different and much slower thing.

Second, small query sets are lumpy. A month of zeros across twenty questions is ordinary for a young site and is not evidence that anything is broken; we treat the reading as useful at six to eight weeks.

You can run the same style of probe on your own domain with the free AI Visibility Checker, or check whether assistants can reach your pages at all with the AI Crawler Check.

That it is a ranking with positions. There is no fourth place inside an answer. The unit is a share of a query set over repeated runs, not a place in a list.

That one check proves something. Assistants sample. The same prompt asked twice can return different sources, so a single good run and a single bad run carry the same weight: none.

That mentions and citations are one thing. Being named in the prose, being linked in the source list, and being paraphrased with no credit at all are three different outcomes with three different fixes.

That publishing more raises it. Volume raises page count. Answers get lifted when a page states a specific claim in a checkable sentence and carries something — first-party data, a real measurement, a datasheet figure — that four other sites do not have.

That it replaces indexing work. Most assistants that browse start from a search index. A page that is blocked, unindexed, or only renders once JavaScript has run holds a citation rate of zero for reasons that have nothing to do with the writing.