When someone asks an assistant "what is the best applicant tracking system for a small team?", the answer names a handful of products. If yours is not among them, you never find out. There is no impression count, no rank, and no referrer in your analytics.
Measuring that gap is what Citerva does. It is worth being precise about what the resulting numbers mean, because they are easy to over-read.
Mention rate
Mention rate is the share of successful answers that named your brand.
The important half of that sentence is successful. If a provider errored, timed out or was never configured, that run is not counted in the denominator. It would be trivial to make the number look better by treating a failed request as "did not mention you", and completely wrong — a request that never happened is not evidence of absence. Every rate in Citerva is shown with the count it came from for this reason.
Share of voice
Share of voice is your slice of the brands named across the answers.
Each answer contributes one point per brand it names. An answer that names you and two competitors gives you one point out of three — a 33% share of that answer, not 100% for appearing in it.
This makes share of voice a share of the set of competitors you track. Add a competitor and the denominator changes; remove one and it changes again. Figures measured before and after such a change are not comparable, which is why the trend chart marks the discontinuity instead of drawing a line through it.
Citations
Some assistants cite sources. Where they do, the cited domains tell you which pages a model reached for when it answered a question about your category.
A citation to your own domain is worth separating from a citation to a review site or a competitor. That distinction depends on Citerva knowing which domains are yours, which is why the domain field on a project is more load-bearing than it looks.
The limit worth stating plainly
These models are not deterministic. Ask the same question twice and the answers can differ — in wording, in which brands appear, and in what gets cited. One scan is a sample, not a measurement of a fixed quantity.
That has two consequences. A single scan is evidence to be read alongside the raw answer, not a verdict to be quoted. And a change between two scans is not automatically a change in the world — it may be variance.
It also means nobody, Citerva included, can promise that a specific change to your site will make a model recommend you. What the data can do is narrower and more useful: show you which questions you are absent from, which competitors are named instead, and which sources those answers lean on. What to do about it is a judgement, and it should be made while looking at the actual answer.