AI visibility checker

Ask the question your buyers ask. The checker puts it to Claude, OpenAI and Perplexity with live web search switched on, then shows whether they name you, whether they cite your site, and who they name instead.

Free, no emailLive web searchNames stay in your browser

Ask the engines

Write the question the way a buyer would type it. Only the question is sent, with the country and engines you pick. Your brand, site and competitors are matched against the answers on this page.

0 / 400
Engines

Separate other spellings of your brand, and your competitors, with commas. A competitor written as a domain is also checked against the cited sources.

What the checker does

Search is not the only place buyers ask for a shortlist any more. They ask an assistant, and the assistant searches the web, reads a handful of pages and answers with names. If yours is not one of them, you are not on the list, and no ranking report will show it.

The checker sends your question to each engine switched on above, through its API, with that engine’s own web search turned on. It adds no instructions, so each model answers the way it would answer anyone. As the answers arrive, this page reads them for four things:

  • Named. Whether your brand appears in the answer, how many times, and how early.
  • Cited. Whether your site is one of the sources the answer cites, or was read and passed over, or never came up.
  • Who else. Where each competitor you entered is named, and in what order.
  • Which sources. Every domain cited, across all the answers.

Reading the results

Named and cited are different wins. An engine can recommend you while citing a review site that mentions you, or cite your page for a fact without recommending you at all. The first is visibility; the second is authority without the sale.

Read, not cited, is its own diagnosis. Where an engine reports every page it opened, the checker separates the pages it read from the pages it cited. A site that was read and passed over was found, and judged less quotable than something else. That is a content problem, and a different one from not being found at all.

Order matters in a list. Most shortlist answers put someone first. The table shows the order of first mention in each engine for you and every competitor you entered.

The sources table is a target list. When a directory, a trade publication or a forum thread is cited by every engine, that page is carrying the answer. Earning a place on it is usually easier than outranking it, which is why the sources table matters more than any single answer.

What it cannot tell you

  • One question is one sample. The same question can get a different answer tomorrow, or in the next minute. Read a single result as a reading, not a ranking. My SEO and AI visibility audit runs 50 to 100 buyer questions per market for exactly that reason.
  • API answers are not app answers. ChatGPT, Claude and Perplexity apps add their own instructions, memory, personalisation and location. The checker sees none of that, which makes the reading cleaner and less personal than what any one user sees.
  • The market is approximate. The country is passed to each engine’s search as an approximate location. An engine that does not accept the country answers without one, and the result says so.
  • No Gemini, and no Google AI Overviews. Google’s terms for Gemini answers grounded in Google Search forbid analysing those answers or mixing other content into them, which is exactly what this checker does, so Gemini is left out. AI Overviews and AI Mode are part of Google Search itself, which the checker does not scrape. What Search Console now reports about them is in tracking AI Overviews.

What happens to what you type

This tool cannot run entirely in your browser, because the engines have to be asked from a server that holds the API keys. So the question, with the country and engines you pick, goes to this site’s server, which sends it to the engines switched on — Anthropic, OpenAI and Perplexity — under each provider’s API terms. This site records the question, the engines you picked and the market, with no identifier attached, and the request to OpenAI is sent with storage switched off. Keep personal or confidential details out of the question. Send Do Not Track or Global Privacy Control and nothing is recorded.

Your brand, website and competitor names never leave your browser. The matching happens on this page, which is also why editing a name updates the results without asking the engines again.

To enforce the daily limit, the server keeps a counter against a salted, one-way hash of your IP address. The salt changes every day, and earlier days’ counters are deleted, salt included, when the next day’s first check comes in. The privacy policy has the details.

Once you know where you stand, the question is what to change. Answer engine optimisation for multi-market brands covers what gets a page cited, and the note on llms.txt covers what that file does not do.

Questions

Not exactly. The checker uses OpenAI’s API, not the ChatGPT app. The app adds its own instructions, memory and account history, and may run a different model. Treat the result as a clean reading of what the model says to someone it knows nothing about, and expect it to vary from day to day.
Each engine runs its own searches, reads different pages and weighs them differently, so the same question produces different sources and different names. The sources table shows which pages carried each answer, and that usually explains the difference.
A few a day per visitor, plus a fixed number a day across all visitors, because every check is paid for per question. The form shows how many you have left. When the day’s checks run out, send me the question and I will run it by hand.
Get onto the sources the engines already cite for your question, and make your own pages easy to quote. The sources table shows which sources those are for one question. The audit repeats that across 50 to 100 buyer questions per market and turns the result into a list of targets.

Sources