How do I check what ChatGPT says about my business?

The two minute version: three questions to ask, in order

Ask these three, in this order, in a fresh chat, swapping in your own category and your own city or suburb.

1. What are the best [category] in [city]?
2. Who is the most trusted [category] in [city]?
3. Which [category] in [city] do people recommend most?

Those three are not arbitrary. They are the deterministic fallback set our own scanner uses when it cannot generate better questions for a cohort, and they are published in the pipeline write-up on The Receipts Engine. They work because they are how a person actually asks, and because none of them contains a brand name.

Use a temporary chat, or one on an account that has never discussed your business, if you can. An assistant that has been told who you are all week is not a fair test of whether it volunteers you to a stranger.

What to write down, and the two columns that matter

Two columns: were you named, yes or no, and who was named instead. Everything else is decoration until those two are filled in.

Set up a sheet with one row per question. Record the answer as it came back, not your summary of it, because your summary will quietly soften a result you did not like. If the engine listed six businesses and you were sixth, that is still a yes, and the five above you are the competitive set you actually have in AI, which is often not the set you thought.

A third column is worth adding once you have the first two: what did it cite. If the answer carries links, open them. Those pages are the ones the engine read to write the answer about your category, and they are the pages your competitors are already on.

Why you must leave your own name out of the question

Because a question that names you can only test recall, and recall is not recommendation. Ask is [your business] any good and you will usually get a pleasant paragraph, whether or not the engine would ever have volunteered you to a customer who had not heard of you.

Our scanner enforces the same rule mechanically. Brand names are scrubbed from the question before it is asked: on the per-business path a filter drops any question containing the domain or its stem, and on the cohort path the question generator is only ever given a category and a geography, so there is no brand in the question to begin with. The filter code is published on The Receipts Engine.

The distinction has a name in our data. A business is named when an engine volunteers it unprompted, and recognised when the engine knows it exists once you ask directly but never volunteers it. Across the 137 businesses in our study with complete records from both engines, 26% were recognised and never named. They were not invisible. They were just never the answer.

Do it in both engines, because they disagree

Run the identical three questions in Gemini, because the two engines reach different conclusions about the same business far more often than most people expect.

Across the 137 businesses in our scan with complete records from both engines, Gemini scored higher for 105, ChatGPT scored higher for 7, and 25 were tied. Across all 137 the two engines' scores differed by an average of 25.3 points out of 100. Twenty six of those 137 were named by Gemini and never by ChatGPT, and ten were named by ChatGPT and never by Gemini. Checking one engine and calling it your AI visibility is a coin toss with your own result.

The longer version of that comparison, including why the gap probably exists, is in should I optimise for ChatGPT or Gemini.

What a free scan adds that the manual test cannot

Volume, both engines, and a stored transcript you can go back to. The manual test tells you what happened once, in one chat, on one day, and that is genuinely useful, but it is a sample of one.

Our scanner asks five buyer questions per category and city cohort with the brand name already removed, asks both ChatGPT and Gemini, stores every answer word for word, and counts mentions with a regular expression rather than asking a model to judge. It also checks the site itself: whether robots.txt blocks any of the six AI crawlers we look for, whether there is JSON-LD on the homepage, whether the meta basics pass, and whether an llms.txt exists.

Be clear about what it does not do. It does not measure leads, revenue or rankings, it does not promise a score will improve, and a single scan is still a single day. It is a baseline you can re-run, which is the only honest way to know whether anything you change has worked.

Where to start

The manual test above costs you two minutes and nothing else, and we would rather you ran it than took our word for anything. If you want it done across both engines with the transcripts kept, the scan is free and does not ask for an email.