Free tools / ai-visibility-checker
AI Visibility Checker
in buildWhether ChatGPT and Gemini name your brand when someone asks your category’s buying questions — separated into cited, mentioned, missing and not measured, and which competitor gets named in your place.
Status
the check is in build · nothing on this page runs oneThere is no field to type your domain into, because the check is not finished.
The engine behind this page exists and is tested offline, but it has never been run against a live answer engine. Until it has, a form here would be theatre: it would take your domain and give you back either nothing or a guess, and a guess about whether an AI names your brand is exactly the number this whole product refuses to print.
So this page does three things instead. It sets out precisely what the check will report, so you can judge the method before you trust the output. It explains why a single run is not an answer. And it sends you to the tools that do work today — which fix the inputs an engine reads, and are worth doing whether or not you ever run a visibility scan.
There is no waitlist form on this page either. There is no backend to receive an email yet, and a form that quietly discards what you type is worse than no form.
when it ships it runs server-side — a browser cannot query an answer engine on your behalf · no scan is published, aggregated or used for training without separate opt-in
What you can check right now instead
4 tools shipped · 3 pages live, check in buildThese four are finished and run today. They do not measure whether AI mentions you — nothing does that except asking the engines — but each one fixes or proves an input the engines actually read.
FAQ Schema Generator
Turns your questions and answers into valid FAQPage JSON-LD. Proves your answers are machine-readable as answers, not just paragraphs.
free · runs in your browserOrganization Schema Generator
States who you are in one parseable block — name, URL, sameAs identities. Entity confusion is a common reason a model names someone else.
free · runs in your browserrobots.txt Generator
Writes every AI agent its own group, so nothing inherits a block by accident. Access is the precondition for ever being cited.
free · runs in your browserllms.txt Generator
Builds the file, with the caveat stated on the page: no major AI vendor has confirmed it reads llms.txt. It is a cheap checkbox, described as one.
free · runs in your browserThree more pages are live, but their checks are in build like this one — each shows a labelled example rather than a result: AI Crawler Checker (which AI agents your robots.txt allows), Render Check (what survives when JavaScript does not run), and the AI Readiness Score (those technical inputs in one pass). Nothing on the shelf guarantees an AI mention — no tool from anyone does.
What the check will report
Exactly one of these four applies to one run of one prompt against one engine. They are four different situations with four different fixes, and merging any two of them is how a visibility number becomes worthless.
01
Cited
Your own domain appears as a source of the answer — the exact host or a true subdomain.
A lookalike like yourbrand.com.example.net does not count.
the only state that proves a page was actually fetched
02
Mentioned
Your brand is named in the answer text, but your domain is not among the sources. The model is repeating you from memory.
different problem, different fix — see below
03
Missing
Neither named nor cited. A real, measured absence — the only one of the four that is genuinely a zero, and the one that carries the competitors named in your place.
a measured absence, not a failed check
04
Not measured
The run did not happen or did not succeed: missing API key, HTTP failure, rate limit, timeout, or the call cap. Every one carries a machine-readable reason.
not measured ≠ zero, in the report and in the denominators
The distinction the dashboards merge
Mentioned and cited are two different problems, and they need two different fixes
A model can name your brand without ever having read your site. It has seen the name during training, in a listicle, in a forum thread, in a competitor’s comparison page — and it repeats it. Nothing was fetched. Nothing links back. That is mentioned. A citation is the opposite evidence: your domain is in the answer’s source list, so a page of yours was actually retrieved and used. That is cited.
Concretely: mentioned but not cited means the engine already believes you belong in the category and simply is not reaching your pages. The work is retrieval — crawler access, pages that render without JavaScript, a page that answers the specific question rather than describing your company, and sources the engine already trusts pointing at you. Cited but not mentioned in a rival’s answer is a different day’s work again.
Missing is a third problem entirely: the engine is not reaching for you at all, and the answer names someone else. That competitor list is the useful part of a missing result — it tells you which pages you are being compared against, in which questions.
Reporting a mention as a citation is the easiest lie in this category to get caught in, because anyone can open the answer and look at the sources. So the two never merge — they survive as separate facts on every single run record, not as one blended score.
Why one run is not an answer
Ask an answer engine the same question twice and you can get two different answers, with two different source lists. That is how these systems work, not a fault in them.
Every figure carries its run count
Results are reported as “cited in 3 of 5 runs” — never as a rank, never as a score out of a hundred, never as a single screenshot of one lucky query. A bare number hides whether it came from one run or twenty, and those are not the same claim.
The denominators are deliberate. Cited, mentioned and missing are counted out of measured
runs, so a failed call cannot dilute a citation rate. Not measured is counted out of
attempted runs, so nothing quietly disappears. A cell with zero measured runs prints
not measured — never 0/0, and never a zero.
What that means for the free scan
The free scan is 10 prompts across 2 engines — ChatGPT and Gemini — at one pass per prompt per engine. So each individual cell is a 1-of-1: a sample, not a rate, and it will say so rather than dressing itself up as a percentage. Repeat passes cost money on every prompt, which is why they sit on the paid tiers along with Perplexity.
Google AI Overviews is not covered at all, on any tier, because there is no API for it. A scraped SERP is a number that breaks silently and cannot be defended, so it is reported as not measurable rather than dressed up as a measurement.
The shape of the output
example result
example data · fictional brand · not a live check| Buyer prompt | Engine | Result | Runs | Named instead |
|---|---|---|---|---|
| Best review management software for contractors? | ChatGPT | mentioned | 2 of 3 measured | — |
| Best review management software for contractors? | Gemini | cited | 1 of 3 measured | — |
| Alternatives to a big-name review tool for small home service businesses | ChatGPT | missing | 3 of 3 measured | Northgate Reviews (3), Kerbside (2) |
| Alternatives to a big-name review tool for small home service businesses | Gemini | mentioned | 1 of 2 measured | Kerbside (1) |
| How much does review management software cost? | ChatGPT | not measured | 0 of 3 attempted | rate_limited — no result for this cell |
| How much does review management software cost? | Gemini | missing | 3 of 3 measured | Northgate Reviews (2) |
| Every prompt in the set | AI Overviews | not measured | not measurable | no API — deliberately excluded, never scraped |
every cell is a k-of-N, never a bare number · not measured is counted out of attempted runs, so a broken integration cannot render as a competitor win
Fair questions.
These five answers are also this page’s FAQPage markup — one source, nothing hidden in the schema that is not on the page.
Can I run a visibility check on this page right now?
No. The check is in build and this page will not pretend otherwise — there is no form, because a form here would either return nothing or return a guess. The page is live because it is linked from across the site and a dead end is worse than an honest one. Meanwhile the four generators on the free tools shelf are finished and fix the inputs an engine actually reads.
What is the difference between mentioned and cited?
Mentioned means the model named your brand in the answer text. Cited means your own domain appears among the answer’s sources, which is the only evidence that a page of yours was actually fetched. A model can name you from memory, from a listicle or from a competitor’s comparison page without ever reading your site — so the two are separate facts and they never merge into one score. They are also two different fixes: being mentioned but never cited is a retrieval problem, not an awareness problem.
Which AI engines will the free scan cover?
Ten prompts across two engines: ChatGPT and Gemini. Perplexity sits on the paid tiers, along with repeated passes for averaging. Google AI Overviews is not covered on any tier, because there is no API for it and a scraped result page is not a measurement we could defend.
Why will this not give me a visibility score or a rank?
Because answer engines vary run to run, so a rank would be a fiction with a decimal point on it. Results are reported as a count out of the runs that actually happened — “cited in 3 of 5 runs” — with the competitors named in your place when you were missing. That number can be re-run next month and compared; a score out of a hundred cannot be checked by anyone, including us.
Why is a failed check reported as “not measured” rather than zero?
Because they are not the same fact. Zero means the engines were asked and did not name you. Not measured means nobody asked — a missing key, a rate limit, a timeout, a call cap. Collapsing the two makes a broken integration look like a competitor win, and it is the single easiest way for a measurement product to publish something false. Every not-measured run carries the reason it did not run.
The check is in build. The fixes are not.
Crawler rules, schema and rendering are the inputs an engine reads before it can cite anyone. Four tools on the shelf do that work today, in your browser, with no signup.