Three tools, three different numbers, none of them client-ready.
When a client asks whether they show up in AI answers, the honest problem is not finding a number — it is that the number moves for reasons nobody can name. Kovmere freezes the question set per site, prints every prompt verbatim, and states the range around each rate. This month and last month are then the same measurement, which is the only condition under which a slide is worth putting in a client report. $29 for one client, no account, no call.
- One-time, nothing renews
- No account created
- Back in about 20 minutes
Or start free on a client's domain
Up to 10 questions, 3 runs, one engine, no account and no card — enough to see who the engines are naming in their category.
Free tells you who is being recommended. It does not tell you which pages were cited, whether the engines can fetch theirs, or what to fix — that is the $29 Check. See what the Check adds.
Why the numbers stop moving on their own
Three design choices, all of them checkable on the report itself.
Up to 40 buyer questions are generated once for a site and then pinned. A second scan of the same site asks the identical questions, so a change between two reports is a change in the answers, not a change in the questions. The report prints the set's own version and when it was frozen.
Verbatim, with the model id, the search tool and the run time. Your client can re-ask any one of them and see what we saw — which is a different conversation from "our tool says 14%".
A rate over 40 questions is an estimate, and the report says how wide it is, plus the size of change that this many questions could actually detect. It is not a cutoff; it is the sentence that stops a 2-point wobble becoming a slide.
Each engine's gates come from that engine's own published crawler documentation, tagged official, observed or inferred, and every page is fetched twice — once with the real crawler user-agent, once as a browser — so a firewall that answers one and refuses the other is visible. A crawler-user-agent fetch shows what a crawler would receive; it is not proof that the engine crawled the page, and the report says so beside the result.
Two ways agencies use it
On a client, monthly
One report per client on the same frozen questions: who was recommended, which pages each engine cited instead, and which of their own pages could not be fetched. The fix list is ordered by what the closest page is missing — a price, a number, a comparison, steps — so it hands your team tickets rather than a score.
On a prospect, before you pitch
"Here are the questions your buyers ask, here is who the engines recommend instead of you, and here is why three of your pages could not be read." It is a shareable report on their own domain, for $29, with no account for them to create.
We do not promise citations or rankings — only that the measurement is reproducible and that you can check every number on it.
What we sample, and why it is an API
Samples come from the OpenAI web-search API (the model id is printed on the report) and the Perplexity Sonar API — not the consumer ChatGPT or Perplexity apps, whose answers also depend on memory, location and personalisation.
For a client report that is the whole argument. An API sample is reproducible: it carries no personalisation, no memory and no location, it prints the model id and the search tool it used, and the identical frozen questions can be re-asked thirty days later and compared. A reading that depends on whose account asked it cannot be any of those things. How we measure.
Start with one client, for $29
Up to 40 buyer questions × 3 samples, the full page-verified source ledger, per-page reachability for 6 pages, and a fix list. One-time — nothing renews, and no account is created. If it earns a slide, run the next client.