Does ChatGPT recommend your business? How to check, properly

When a buyer asks ChatGPT for a business like yours, does it say your name? You can get a rough answer yourself in half an hour. This is the procedure we use professionally, minus the paid tooling — and an honest note on the two ways a DIY check misleads.

Step 1: write the questions a buyer would actually ask

Not "what do you know about [your name]" — that question tells the model who to look for. Write questions from someone who has never heard of you:

• Someone working out what exists: "What is [your category] and how do I choose one?"
• Someone comparing: "Best [category] in [your city] according to reviews"
• Someone choosing: "Which [category] should I use for [specific problem]?"
• Someone about to buy: "How much does [service] cost in [city]?"

Ten questions is enough for a first reading. Keep them in a file — you will ask the same ones again next month, and comparing runs only works if the questions are identical.

Step 2: ask in clean sessions

One question per new chat. Anything you asked earlier in the same conversation contaminates the next answer. Do not mention your business in the question — the point is to learn whether the assistant produces your name on its own.

Step 3: record the whole answer, not just yes/no

"Did it mention me" is the least useful part. What you want to know:

• Was it recommended ("I'd go with X"), mentioned in passing ("others include X"), or cited as a source while someone else got the recommendation?
• Which businesses were named instead, in what order?
• What reason did the assistant give — price, reviews, a specific service page?

The difference matters more than the count. A business named ten times as a footnote is in a weaker position than one named three times as the recommendation.

Step 4: ask the same questions again another day

This is the step most DIY checks skip, and it is where the honest answer lives. Assistants that search the web do not give identical answers twice; in our own measurements, each key question is asked three times for exactly this reason. One run is an anecdote. Two runs on different days is the beginning of a measurement.

The two ways DIY checks mislead

First, string matching flatters you. If the answer contains your name, a search says "mentioned." But "named in passing" and "recommended" are different outcomes with different value, and no search box can tell them apart — it takes a reading of the whole answer against a rubric. Most free tools do the search-box version, which is why their scores look better than your leads do.

Second, one session on one day is not a rate. If 4 of your 10 questions came back with your name, the honest reading is "4 of 10 on that day," not "40% visibility." The percentage only starts meaning anything when the same questions have been asked across several days and the answers stop surprising you.

What DIY cannot tell you

• Why the assistant names or skips you — which sites its answers were built from, and whether any of them mention you at all
• Whether the facts it states about you are right — including whether the assistant can confirm the business exists at all
• Whether the answer holds on Google's own results page (AI Overviews), which runs on different machinery than the chat

Those gaps are exactly what a measured AI visibility audit fills: the same questions across every surface, repeated for stability, every answer graded against a published rule, and the full text kept so you can check any claim. You can read a complete real report before deciding whether to pay for one — and the free check shows you the questions first.