AI citation tracking: Who gets named.
Who gets cited.
And the answer to prove it.

A mention means the answer wrote your brand name in its prose. A citation means the answer linked a page on your domain. They are recorded as two separate flags on the same answer, so an answer can be one, both or neither. Mentioned but not cited is the interesting case: you are in the conversation, but the assistant is not pulling from your site to say it. The same answers, grouped by subject, are on the topics tab, and the free AI content readiness checker shows whether a page of yours is easy to quote.

A case-insensitive word-boundary match against the answer text, over your brand name, any aliases you store, and the first label of your domain. Verand also derives a small set of aliases so a restyled name is not scored a miss: symbols spelled out both ways, punctuation flattened to spaces, and a trailing legal or generic suffix trimmed, but never down to a single word. That last rule is deliberate, because a one-word alias would match every other firm with the same surname and overcounting is worse than undercounting for a product that sells measurement.

Yes, and cite rate can read over 100%. The two are counted independently over the same set of answers, so an assistant that links your page without writing your name adds a citation and no mention. It happens in practice: on one fourteen-day window of Willowdale's data, one assistant returned two answers, neither of which wrote the name and one of which linked a Willowdale page.

Both. For ChatGPT and Gemini, Verand takes the assistant's own source list and the markdown links written into the answer, and deduplicates them by URL. For Perplexity and Claude it takes the annotations on every part of the answer. For the two Google surfaces it takes the references attached to the AI answer. The match is on the registrable domain, so a blog subdomain of yours counts.

It depends which kind of nothing it is, and the two are never merged. Google renders no AI Overview for plenty of queries; your brand could not have appeared in an answer that does not exist, so nothing is recorded and the count is simply smaller. A provider error or a missing credential is a failure, and that is written to its own table with the provider's own message, so a broken engine can never be read as a day you were not mentioned.

Not by default. The Unbranded toggle is on, and it excludes any prompt whose own text names you, because an answer to “is Willowdale Equity legitimate” mentions you by construction and measures nothing. Turn it off and those prompts come back in, tracked as reputation rather than visibility.