Step-by-step guide Get Recommended by ChatGPT: local service businesses →
To check whether ChatGPT recommends your business, open a temporary chat, ask six questions a customer would ask without naming yourself, ask each one a second time, and count how many of the twelve answers name you. It takes about 20 minutes and no paid tool. What you get is a first reading, not a verdict, because ChatGPT writes a fresh answer every time and the list of names moves from one run to the next. Below is how to run that one sitting properly, the three things that quietly colour the answer, and how to read what comes back.
Why asking once tells you almost nothing
A single answer is one draw from a hat. If your name comes out, you've learned little. If it doesn't, you've learned just as little. Two studies from this year put numbers on that.
- Local answers. On 16 September 2026, BrightLocal published a study of about 200,000 local AI searches. When the same question was asked again in ChatGPT, only 23% of the businesses named were the same. A business that appeared once came back in about half of later runs. The study's own summary: "71-80% of businesses appear in half the attempts or fewer, so inconsistent inclusion is still expected."
- Brand and product answers. On 28 January 2026, SparkToro and Gumshoe reported what happened when 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI 2,961 times. The chance of getting the same list of brands twice from ChatGPT was less than 1 in 100. The authors say plainly that they are not professional researchers and invite more rigorous work. Their conclusion still fits the local data: a share of answers across many runs is a reasonable thing to measure, and a "ranking position" is not.
We saw the same thing ourselves. On 9 November 2026 we asked ChatGPT the same question about a physiotherapist for a running injury in Amsterdam-Oost twice, minutes apart. Its first choice changed. The full test is written up in how ChatGPT picks local businesses.
So the method is simple: don't look at an answer, count across answers.
Three things that colour the answer before you type
A test is only useful if it shows what a stranger would see. Three things get in the way.
- What the assistant already knows about you. If you've ever discussed your own business with ChatGPT, a normal chat is not a neutral place to test. We used Temporary chat, the switch at the top of a new chat, which skips memory and isn't saved to your history. Google has the same issue: its AI Mode help page says personalisation applies if you're 18 or older with history and personalised recommendations switched on. Sign out or use a private window.
- Where the assistant thinks you are. Local answers lean towards nearby businesses. In BrightLocal's study, 53% of the businesses ChatGPT picked were within 5 km of the searcher, and every check was localised to the business's own location. You can't easily move your computer, so put the place in the question: "in Amsterdam-Oost", "near Oosterpark". That's how customers ask anyway.
- How the question is worded. The SparkToro study also found that people phrase the same need in very different ways. A question written by the owner tends to use the owner's vocabulary. Write yours the way customers say it at the front desk or in an email, and never include your business name.
The 20-minute test
- Write six questions. Two that describe a problem in the customer's words ("My lower back hurts from sitting at a desk all day. Who can help in Amsterdam?"). Two that name a service ("Which practices in Amsterdam offer dry needling?"). Two about place and practical needs ("English-speaking physio near Oosterpark with evening appointments").
- Open a temporary chat for each question. A new chat every time, so one answer can't colour the next. Check that the answer comes with sources. If it doesn't, ask it to search the web and answer again, so that you're testing what it can find today.
- Write down every business named, in order. A spreadsheet with one row per question is enough.
- Score each answer. R if you're recommended, meaning named first or with a reason to pick you. M if you're mentioned somewhere in the list. A dash if you're absent. Note who came first when it wasn't you.
- Ask all six again. Same wording, fresh temporary chats. Score them the same way. You now have twelve answers.
- Count. Answers that named you, out of twelve. Then, separately, answers that recommended you.
If you have ten minutes left, ask the same six questions once in Google's AI Mode in a private window. It draws on different sources, and in BrightLocal's study ChatGPT's picks overlapped with either Google surface by only 19 to 20%.
How to read your result
Twelve answers is a small sample, so read patterns, not decimals. For scale: the businesses tracked in BrightLocal's study were named in 33% of ChatGPT answers on average, which would be 4 of your 12. Treat that as context, not as a pass mark, because those businesses and their markets are not yours.
| What you see | What it most likely means | Where to look next |
|---|---|---|
| Never named in twelve answers | ChatGPT isn't finding a reason to pick you for these questions | Whether its crawler can reach your site, and whether your pages state what you do, for whom and where |
| Named now and then | You're in the pool. This is the normal state for most businesses | Which question types name you and which never do |
| Named, but never first | Competitors state specifics that you don't: specialisms, languages, opening hours | The reasons ChatGPT gives for the business it puts first |
| Named for services, absent for problems | Your site lists treatments or services but not the situations customers describe | The wording on your service pages |
| Named with wrong hours, address or phone | An old listing somewhere still carries the old details | The source ChatGPT names when you ask where it got them |
For the first row, start with whether your website is blocking ChatGPT's crawler. For the last, see which business listings AI assistants read and how to make them match.
Ask why, and where from
The count tells you where you stand. Two follow-up questions tell you why. Pick one or two answers where a competitor came first. In the same chat, ask why it recommended that business first, and then which websites it used for the answer.
In our Amsterdam test the reasons were strikingly concrete. One practice won the running-injury question because a therapist's profile said they specialise in running injuries. Another won the desk-related back pain question because its page said it treats lower-back pain from prolonged sitting. Reviews came in as supporting reasons, with star ratings and review counts quoted back.
Treat the explanation as a lead, not as proof. The reasons are generated text too. Open the pages it names and check that they say what the assistant claims. The list of websites is the more useful half: every site it read where you're missing or out of date is somewhere you can do something about.
A second check: what does it say about you by name?
Everything above tests whether you're recommended to someone who has never heard of you. A different question is whether ChatGPT has your facts right. For that, break the no-name rule once: in a fresh temporary chat, ask for your opening hours, address and phone number by business name and city, and ask which sources it used.
If a detail is wrong, the cause is nearly always a listing you forgot about. Fix the source it names, then check that your website and your main profiles say the same thing. A wrong phone number costs more than a missed mention, because it reaches the customers who already chose you.
What one sitting can't tell you
- It covers one assistant. In BrightLocal's consumer research of 10 March 2026, 31% of consumers had used ChatGPT for local business recommendations in the past year and 23% had used Google's AI Mode. Your customers are spread across both, and the two rarely name the same businesses. In our own test, the two shared only two practices for one question about sports physiotherapy.
- It covers one day. Assistants re-read the web continuously. A sitting next week will differ.
- Twelve answers can't show a small change. Going from two mentions to four is inside the noise. Only compare averages with averages.
- A checker tool has the same limits. Many websites offer an instant AI visibility score. Before trusting one, ask how many runs are behind the number. One run per question is the same as asking once yourself.
The honest use of a first reading is to decide whether there's a problem worth an afternoon. If you were never named in twelve answers, there is.
From a first reading to a number you can track
A proper measurement needs more questions, more assistants and more days. That's what our guide, Get Recommended by ChatGPT, local service businesses edition, sets out: twelve questions, three assistants and three runs across one week, with a printable score sheet, the exact follow-up prompts, a source-gap list and a 30-day plan for fixing what the audit finds. It can't promise that any assistant will recommend you, and nobody honestly can. It shows you how to measure where you stand and whether your work moved the number.
Whatever your first count is, write it down with the date. Four weeks from now you'll want something to compare against. Neurova AI builds AI systems to healthcare standards from Eindhoven, and visibility work like this is part of what we do for practices and service businesses. If you'd rather hand the whole check over, tell us what you need.
Frequently asked questions
How do I check if ChatGPT recommends my business? Open a temporary chat, ask six questions a local customer would ask without using your business name, then ask each question a second time in a fresh chat. Write down every business named and count how many of the twelve answers include you. Count across runs instead of judging one answer, because the names change each time.
How many times should I ask ChatGPT the same question? At least twice for a first reading, and more if you want a number you can track. In BrightLocal's September 2026 study, repeat runs of the same local question in ChatGPT shared only 23% of the businesses named, and a business that appeared once came back in about half of later runs. Our guide's full audit uses three runs across one week.
Why use a temporary chat for the test? So that your earlier conversations cannot shape the answer. If you have ever talked to ChatGPT about your own business, a normal chat may not show what a stranger would see. In our test on 9 November 2026 we used Temporary chat, which skips memory and is not saved to your history.
Does ChatGPT recommend the same businesses to everyone? No. Each answer is generated fresh, and the wording of the question and the place named in it both change the result. A January 2026 SparkToro and Gumshoe study of 2,961 runs found less than a 1 in 100 chance of getting the same list of brands twice from ChatGPT.
Do I need a paid AI visibility tool to check this? Not for a first reading. A browser, a spreadsheet and 20 minutes are enough. A tool earns its place when you want many repeat runs without the typing. Before trusting any score, ask how many runs are behind it, because a score built on a single run has the same weakness as asking once yourself.