Expert Insights

How to check whether AI knows your business

You can find out in twenty minutes whether AI assistants know your business exists, what they think you do, and who they name instead of you.

22 September 2026

People ask assistants for recommendations now, in the same way they used to ask search engines. The useful thing about this is that you can check what they say about you, and unlike search rankings, the answer is written in plain sentences.

Most businesses have never looked. It takes about twenty minutes and it is worth doing before you spend anything on being more visible, because it tells you which of three problems you have.

Set the test up so the answer means something

Four rules. Skip these and you will get a flattering answer that tells you nothing.

Log out, or use a private window. An assistant that has your history knows your business because you have talked about it. A stranger asking the same question has none of that. You want the stranger's answer.

Start a fresh conversation for each question. Once a name appears in a conversation, the assistant will happily keep using it. That is memory, not knowledge.

Use the same wording every time. Write your questions down and reuse them exactly. If you reword them next month, you cannot tell whether something changed or you just asked differently.

Write the answers down with the date. The point of the exercise is comparison over time. An answer you did not record is an answer you cannot compare to.

The questions to ask

Ask all four of these, in four separate fresh conversations, on each assistant you test. ChatGPT, Gemini and Perplexity are a sensible set, and Google's AI Overviews should be checked by simply running the searches in Google.

One. The direct question. "What is [your business name]?" This tells you whether you exist to the system at all, and whether it has you confused with somebody else who shares your name.

Two. The category question. "Recommend a [what you do] in [your city]." This is the one that matters commercially, because it is closest to what a real buyer types. Note every business it names and the order.

Three. The problem question. Ask about the problem your customer has rather than the service you sell. Somebody does not wake up wanting brand design, they wake up thinking their business looks unprofessional next to competitors. Ask it the way they would.

Four. The comparison question. "What are the differences between [your business] and [a competitor]?" This reveals what the system believes about you, which is often more revealing than whether it mentions you.

How to read what comes back

There are three outcomes and they call for different work.

It does not know you exist. You get a generic answer, or a polite statement that it has no information. This is the most common result for a business with a young website and no established profiles, and it is not a ranking problem. There is simply not enough about you in the places these systems draw from. The fix is creating that material, starting with the sources they trust most, which are your own site, your own profiles and independent mentions of you.

It knows you but describes you wrongly. Wrong city, wrong services, merged with another business of a similar name, or an old description you have outgrown. This is usually the easiest of the three to improve, because the raw material exists and the problem is clarity. It normally traces back to your own pages saying it unclearly, or to an outdated listing somewhere you forgot about.

It knows you and names somebody else anyway. The hardest case, and common. We have written about that one separately, because the reasons are specific.

What not to conclude

Two warnings, both of which matter.

These answers are not stable. Ask the same question twice and you can get different businesses named. That is how these systems work, and it means one run is an anecdote. The pattern across several runs and several weeks is the finding.

And nobody controls this. Any agency, including ours, that promises to get you named by ChatGPT is selling something it cannot deliver. What can be done is to make the information about you clearer, more complete and better supported in the places these systems read, and then to measure honestly whether it moved.

Turning it into something you can track

Keep a simple sheet. One row for each question, one column for each assistant, a new set of columns each month. Record whether you were mentioned, in what position, what it said about you, and which competitors appeared.

After two or three months you have something almost nobody in this market has, which is a record of what the systems your customers are starting to use actually say about you, and whether the work you are doing is moving it.

If you would rather have that done properly, the free deep digital audit includes this test across four assistants, run logged out in clean sessions, with the wording fixed so the results can be compared month to month.

We run this test properly, for free.

The free deep digital audit runs these checks across four assistants, logged out, in clean sessions, with the results written down and dated so next month can be compared to this one.

Questions

How do I check if ChatGPT knows about my business?

Log out or use a private window, start a fresh conversation, and ask what your business is. Then ask it to recommend a business in your category and city. Use the same wording every time and write down the answers with the date.

Why do I get a different answer every time I ask?

These systems do not give fixed answers. One run is an anecdote. Ask the same fixed questions several times across a few weeks and look at the pattern rather than any single reply.

Why should I log out before testing?

Logged in, the assistant may know your business because you have discussed it before. That is memory, not public knowledge. A stranger gets a different answer, and the stranger's answer is the one that matters.

Can an agency guarantee my business will be recommended by AI?

No, and you should be careful with anyone who says otherwise. Nobody controls these outputs. What can be done is to make the information about you clearer and better supported in the sources these systems read, then measure whether it moved.