You can do this yourself in about fifteen minutes, for nothing, and you should do it before you pay anybody. Here is the method we use, minus the software.
Step one: ask the customer's question, not yours
The instinct is to type your business name. Don't. An assistant asked about a named business will usually say something polite, and it tells you nothing about whether strangers are being pointed at you. Ask what a customer with a problem would ask: "who should I use for emergency boiler repair in Leeds", "best storage for students in Bristol", "which firm should I use for probate in Newcastle".
Step two: ask it more than once
This is the step everyone skips and it is the one that matters. Ask the same question three or four times, and vary the wording a little between attempts. You will very likely get different names. Across the UK markets we measure, identical questions asked minutes apart produce a materially different shortlist about 19% of the time. If you check once and see yourself, you have not established that you are visible. If you check once and do not, you have not established that you are invisible.
Step three: write down the right things
- The question, word for word.
- The date, and which assistant you used.
- Every business named, in the order they appeared.
- Any sources or links it cited.
That last one is the most useful and the least obvious. The sources tell you where the machine goes for opinions about your trade, which is where your evidence needs to exist.
Step four: check the other places too
ChatGPT is not the only one. Google's AI Overviews sit above the ordinary results for many searches, and Perplexity behaves differently again. They disagree with each other more than people expect. It is worth running your two most valuable questions through each.
How to avoid fooling yourself
- Do not use your own name as the test.
- Do not stop at one run in either direction.
- Log out or use a fresh window, so your own history is not flattering you.
- Repeat the exercise in a month, using exactly the same questions, or the comparison means nothing.
When to bring someone in
Do it yourself first. If you find you are consistently absent from questions that matter, or that the assistant is saying something about you that is out of date or wrong, the next questions are which sources are causing it and what changes it, and those need a wider and more repeatable measurement than a person with a browser can run. If you would like ours, we run ten of your questions before we ever meet, at no cost. If it turns out you are fine, we will say so.
Common questions
How do I check if ChatGPT recommends my business?
Ask it the question your customer would ask, not your business name. Something like 'who should I use for [service] in [town]'. Ask it several times in slightly different wording, on different days, and write down who is named each time. Searching your own name will nearly always produce something flattering and tells you almost nothing about whether you are being recommended to strangers.
Why do I get a different answer every time?
Because these systems are not deterministic and they retrieve fresh sources each time. In our measurements, identical questions asked minutes apart return a materially different set of businesses about 19% of the time. This is normal, and it is why a single screenshot is not evidence.
Is there a tool that just tells me my score?
Several, and they are worth exactly what they measure. A score is a summary of somebody's chosen questions, asked some number of times, by a method you usually cannot inspect. Before you trust a number, ask which questions it used, how many runs it averaged and whether you can see the answers.
What should I write down?
The question exactly as you asked it, the date, which assistant, who was named and in what order, and any sources it cited. That record is what lets you tell in two months whether anything changed.
Updated August 2026