What AI visibility actually measures
AI visibility is the share of relevant answers in which your brand is named. It is not a ranking, it does not come with a position number, and it cannot be derived from one. An assistant composing an answer is choosing what to say, not ordering a list.
Three numbers come out of a run, and they answer different questions. Mention rate: across your prompts and engines, how often were you named at all. Position: when you were named, where you came among the brands the answer lists, because being named fifth is not the same result as being named first. Sources: which URLs the engine actually cited on its way to that answer.
The third is the one that turns measurement into work. The first two tell you where you stand; the citation list tells you which pages the engine is reading, which is where any change has to start.
Why the number moves when nothing changed
These models search the live web and are non-deterministic. The same prompt can name you today and skip you tomorrow with no change on your side and none on theirs. Any AI visibility tool that reports a single confident number is hiding this, not solving it.
The honest response is to treat one run as one sample. That is why runs are stored rather than overwritten, why the same prompts are scored for you and your competitors in the same run, and why the interface is built around comparison over time instead of a headline figure.
Read a single check as a snapshot with real error bars. Read six weeks of checks as a trend. The trend is the thing that is actually true.
How the answers get built, and what that means for you
Four of the five engines run through DataForSEO's AI Optimization API, which calls each provider's own model with web search enabled. That detail matters more than it sounds: it means a miss is informative. The model searched, read pages, and still did not name you, which is a different and much more useful finding than a model answering from memory with a two-year-old training cutoff.
The Google AI Overview is the exception, read from live SERP data because there is no API for it. Nothing here scrapes the ChatGPT app, and we would rather say what the pipeline is than let 'we monitor ChatGPT' do work the implementation does not.
Practically, answer engine optimization comes down to being present and quotable in the sources these engines read. That is why this sits in the same product as the site audit and the editor rather than as a standalone dashboard: the gap shows up here, and the page that closes it gets written there.
What a check costs, and when it runs
A run uses one AI visibility check per buyer question per assistant, and the app says how many before you start it. Every plan includes a monthly allowance of checks, and the weekly re-check counts against it like a run you pressed.
Opening the page reads your stored runs and spends nothing; a new run happens when you press the button. Once a week the app also re-asks your last check's questions, so the trend keeps itself without you, and you can turn that off per workspace. Weekly is the right cadence for most brands: often enough to see a trend, rarely enough that a noisy day does not read as a change.