← Back to Blog

1 Oct 20267 min read

AI visibility check: is ChatGPT mentioning your company?

How to run an AI visibility check: whether ChatGPT, Perplexity, Gemini and Google AI Overviews mention you, why one check misleads, and what changes it.

An AI visibility check answers one question: when a buyer asks ChatGPT, Perplexity, Gemini or Google's AI Overviews the questions your customers ask, does your company come up, and is it cited with a link or only mentioned? You can run one by hand in an afternoon. This post shows how, why a single check misleads, and what actually changes an answer.

It matters because a buyer who used to type a question into Google now often asks an assistant. They ask ChatGPT which vendors handle a problem, ask Perplexity for a shortlist with sources, or read the AI Overview Google puts above its own results. If your company is not in those answers, you are not on the shortlist, and no amount of page-one rankings tells you so.

How to run an AI visibility check by hand

Start with ten questions your buyers really ask. Not your product name, and not your category label. The questions a head of operations types at 6 pm: "how do freight brokers quote after hours", "which tools identify companies visiting a website", "what does a SOC 2 checklist for a startup look like". Your sales calls and your support inbox are the best sources.

Then ask each question, in a fresh session, on four surfaces:

  • ChatGPT, signed out, so no memory of you colours the answer.
  • Perplexity, which shows its sources, so you can see who it cites.
  • Gemini, which draws on Google's index differently from the other two.
  • Google itself, for the AI Overview above the results, if one appears.

Record three things per answer: whether you are named at all, whether you are cited with a link or only mentioned in passing, and who else is named. The third column is the useful one. The companies an assistant keeps naming are your competitors in that answer, whether or not you think of them as competitors.

Why one check does not measure AI search visibility

An assistant writes a new answer every time, so it is not an index. Ask the same question twice and you can get a different shortlist. Ask from a different country and the answer changes again, because the underlying search results change. A single check tells you what one answer said on one day. It does not tell you your visibility.

What you want is a rate: across your questions, across the assistants, across several weeks, how often are you cited? That number moves slowly and honestly. It is the only figure worth putting in front of a founder, because it does not swing with a single lucky answer.

This is why Mati runs the check weekly rather than once. Each week it asks the approved questions on ChatGPT, Perplexity, Gemini and Google AI Overviews in each market you sell in, keeps every answer on file, and reports your citation rate and your share of voice against the competitors those answers name. A check you cannot repeat is an anecdote; a check you repeat is a measurement.

What is actually in an answer

When you read the answers side by side, a pattern appears. Assistants prefer pages that answer the question directly in their first lines, that carry specifics (a number, a step, a named constraint), and that other pages cite. They draw from comparison pages, from documentation, from forum threads and from the odd well-written blog post. They rarely draw from a home page that says "the leading platform for".

So an AI visibility check is also a content audit. For each question where you are absent, look at the pages the assistant did cite. Usually one of three things is true: you have no page on the topic, you have a page that never states the answer plainly, or your page exists but no assistant's crawler can read it because a robots rule blocks it. Each has a different fix.

What improves your AI search visibility

  • Answer first. Put the direct answer to the question in the first two sentences of the page, in plain words, then explain. Assistants quote the sentence that answers; they do not hunt for it.
  • Be specific and quotable. A figure with its source, a named step, a clear condition. Vague pages are not cited because there is nothing to cite.
  • Let the crawlers in. Check your robots file for the AI crawlers' user agents. Blocking them is a legitimate choice, but it is a choice; do not make it by accident.
  • Get named by others. Comparison pages and lists are what assistants read when a buyer asks for options. Being on the right list matters more than your own page for that kind of question.
  • Re-measure. Change one thing, then run the weekly check for a few weeks before you judge it. This is slow by nature. Anyone promising a citation next week is guessing.

How Mati runs it

Mati is an AI that runs your marketing end to end, and its AI answers page is the AI visibility tool inside it, built around exactly this loop. You approve the questions it asks. It runs the check every week on the four surfaces, per market, and keeps every answer. It reports the citation rate, the mentions, the share of voice and the gaps: the questions where you are absent and who is present instead. For each gap it diagnoses which of the three causes applies and proposes the fix, which might be a new article, an answer-first rewrite of an existing page, or a change to who may crawl you. The fix joins the same campaign plan as your SEO articles and LinkedIn posts, so one plan carries the story across every channel. You approve the fix, and the following weeks show the before and after.

None of this promises a citation. It measures, proposes, and measures again, which is the only honest thing a tool can do with an assistant it does not control.

See whether the assistants name you

On a 30-minute call we run your first AI visibility check together and show you the questions where you are absent.

Keep reading