How to check what AI search says about your business

TL;DR: You can run a useful AI visibility check yourself. Start with Google's own data in Search Console, then ask the AI tools your buyers use the questions your buyers ask, a few times each, and write down three things every time: were you mentioned, was your site cited, and who got the credit instead. One answer is an anecdote. A logged sample is a baseline.

A business owner asks ChatGPT about their own company. The answer is friendly and mostly right. They take a screenshot, send it to the team, and call it a win.

Next week a customer asks a slightly different question and gets a competitor.

The screenshot was never a measurement. It was a mood.

Start with the data Google already gives you

For Google's AI features, you don't have to guess. In June 2026 Google announced generative AI performance reports in Search Console, and as of 31 August 2026 they are rolled out to all sites. They show how often your URLs appeared in AI Overviews and AI Mode, which pages, which countries and which devices.

That is the closest thing to a scoreboard you will get, and it is free. If your site isn't verified in Search Console yet, that is step zero.

It is also worth reading what Google says it takes to show up at all. Its guide to generative AI features says a page "must be indexed and eligible to be shown in Google Search with a snippet," and the site must be included in Search generative AI features in Search Console. If a page isn't indexed, no amount of checking AI answers will change that. Use URL Inspection before you worry about anything else.

The same guide carries a warning we agree with: no third-party tool "has access to our internal ranking or AI systems." Tools can save you time. They can't see inside Google.

Then ask the questions your buyers ask

Google's report covers Google. ChatGPT, Perplexity, Gemini and Copilot don't send you a report, so you build a small sample yourself.

Pick five to ten questions. Not your brand name alone, which is the easy win, but the questions a buyer types before they know you exist: "who does X near me", "how to fix Y", "best way to Z for a small team". Keep the wording fixed. Changing it every month makes the numbers meaningless.

Then run each question in each tool your buyers actually use. Run it more than once. AI answers change from run to run, and some tools change more than others. Three runs per question is a reasonable floor for a small business.

Write down three things, every time

For each run, log a row in a spreadsheet:

• Mentioned: did the answer name your business? Yes or no.
• Cited: did it link to your own site? Yes or no, and which page.
• Credited instead: if it cited someone else for the point that came from you, or linked a directory instead of your site, who was it?

Add the date, the tool and the exact question. That's it.

The third column is the one people skip and the one that matters most. Being named without being linked means the AI knows you, but someone else gets the click. That gap is usually fixable: a directory listing that outranks your own page, an old profile with the wrong details, or a page of yours that never states the answer plainly.

Our read from running these checks is that the mention and the citation are two different results, and tracking only one of them hides most of the story.

Report counts, not percentages

With five questions and three runs, you have fifteen answers. Write "cited in 4 of 15", not "27% citation rate". The percentage sounds precise and isn't. One extra citation, 5 of 15, moves it by almost seven points.

Keep the denominator next to every number, and compare the same questions month to month. If something shows up in two checks in a row, take it seriously. If it appears once and vanishes, it was probably noise.

Keep the check boring on purpose

It is tempting to hand this whole job to an AI agent that "monitors your AI visibility". We would wait.

Anthropic's engineering post on building effective agents recommends "finding the simplest solution possible, and only increasing complexity when needed," and warns that agent autonomy "means higher costs, and the potential for compounding errors." A visibility check is a fixed list of steps. Fixed steps are a workflow, not an agent. You want the same questions, asked the same way, logged the same way, every month.

If you automate anything, automate the logging. Keep the reading and the judgment with a person.

Acting on what you find

Most of what a check turns up falls into a few piles.

If you're not indexed, fix that first. Nothing else matters until Google can see the page.

If you're mentioned but not cited, look at who got the link. Update that listing, or make your own page answer the question in its first lines so it is the obvious source.

If a competitor is cited for every category question, read their page. Not to copy it, but to see what it answers that yours doesn't. Google's guide calls the content worth citing "non-commodity": first-hand experience and a real point of view, not a summary of what is already out there.

If the AI says something wrong about you, find where it came from. It is usually a page you can correct.

And if nothing shows up yet, write that down too. A zero is a baseline. Ours started at zero across the board, which is the most honest place to begin.

A month-by-month routine

Once a month, pull the Search Console generative AI report and note impressions by page. Run your fixed questions, three times each, in the tools that matter to you. Fill in the three columns. Compare with last month, using counts.

Then pick one thing to fix. Just one. The check is only worth running if it changes what you do next.