← Blog ChatGPT AI visibility Do it yourself GEO SMEs

Check for yourself: Does ChatGPT mention your business? (5 questions, 15 minutes)

2026-08-03 · Martin Nymann · 7 min reading

Check in a quarter of an hour whether ChatGPT mentions your business: 5 questions phrased as your customers would ask them, a template covering the four question types, and — most importantly — how to interpret the results honestly: a single response is a sample, not a survey.

Key takeaways

  • Ask as a customer, not as the owner: ‘Can you recommend a [sector] in [town]?’ — never ‘Do you know [company name]?’. Start a new conversation for each question, and never mention your own name in the question.
  • Choose five questions from four categories: recommendation, comparison, problem and local — and for each answer, note down whether you are mentioned, who is mentioned instead, and which sources the answer is based on.
  • A single response is a snapshot, not a measurement: the same question might yield a different answer tomorrow. Make a note of the wording, date and model, and only draw conclusions once the pattern has repeated itself.
  • Citations and recommendations are two different outcomes (based on our own survey, 576 AI responses): the technique predicts citations — recommendations also require reputation and publicity.
  • If you weren’t mentioned, start with the basics: the free test takes 60 seconds to check whether the AI engines can even read and understand your site.

You can test this in a quarter of an hour, without any tools at all: ask ChatGPT five questions, phrased as your customers would ask them — not as the owner would. For each answer, note down whether your business is mentioned, who is mentioned instead, and what sources the answer is based on. This guide provides you with the template for the five questions, the pitfalls that can ruin the test, and — most importantly — how to interpret the results honestly: a single response is a sample, not a survey.

Why is it worth testing now?

Because a growing proportion of your customers’ research never reaches your website. Ahrefs’ analysis of 300,000 keywords found a 34.5 per cent lower click-through rate for the top position in Google when an AI answer appears at the top — and 58 per cent in the follow-up measurement in December 2025. When the answer replaces the click, the customer’s shortlist is decided within the answer itself. You’ll only know if you’re on it if you ask.

The good news is that visibility can be improved: the Princeton/Georgia Tech research (KDD ’24) found a visibility boost of up to 40 per cent in generative search engines through targeted content optimisation. But before you optimise anything at all, you need to know your starting point — and that’s exactly what the neighbourhood test gives you.

How do you ask questions as a customer — rather than as the owner?

The most common mistake is to ask, “Do you know [your company name]?”. That answer is virtually worthless: your customers don’t ask for you by name — if they did, they wouldn’t need to ask. Customers are looking for a solution to a problem: “Who can help with VAT for a sole trader?”, “Recommend a plumber in Kolding”, “Which accounting software is suitable for an online shop?”. These are the questions you need to test — because they’re the ones that determine whether you’re even in the customer’s consideration set.

Two practical rules before you start:

  • Start a new conversation for each question. ChatGPT uses what’s already in the conversation as context. If you’ve just mentioned your own company, it will skew the answer to the next question.
  • Never mention your own company name in the question. You’re testing whether the AI can find you on its own — not whether it can parrot what you say.

Which 5 questions should you ask?

Customer questions fall into four categories. Choose five questions so that you cover at least three of these categories — and phrase them in your own words, tailored to your industry and city:

TypeTemplateExample
Recommendation"Recommend a [sector] in [town]""Recommend an accountant in Aalborg for a small online shop"
Comparison"What is the difference between [solution A] and [solution B]?""Fixed fee or hourly rate for a solicitor handling a property transaction?"
Problem"How do I solve [specific problem]?""My underfloor heating system is making a knocking noise — what should I do?"
Local“[Service] near [town/area]”"Dentist offering emergency appointments in Odense C"

How to run the test:

  1. Open a new conversation in ChatGPT (the free version is fine — that’s the one most customers use).
  2. Ask one question from your list, word for word as you’ve written it down.
  3. Make a note of three things: Were you mentioned? Who was mentioned instead? Which sources does the answer link to or refer to?
  4. Save the wording of the question, the date and the model (e.g. “GPT-5, 3 August”) — you’ll need this if you want to compare results later.
  5. New conversation, next question — until all five have been tested.

How do you interpret the answers?

Each answer falls into one of three categories, and each requires a different response:

  • Correctly stated: Your name, service and details are correct. Make a note of the sources the answer draws on — these are the ones you should prioritise.
  • Mentioned, but incorrect: The AI recognises you but provides inaccurate information — an old address, incorrect services, a name that’s almost yours. This is often a sign that your information is difficult for machines to read or points in several directions.
  • Not mentioned: The most important category to interpret correctly. Look at who was mentioned: Are they nationwide portals? Competitors with strong review profiles? Directories and overview pages? The pattern in who the AI chooses instead tells you what sort of sources it relies on in your industry.

And here’s a truth that very few people will tell you: there’s a difference between being cited as a source and being recommended as a business. When we analysed 576 AI responses (our own survey, July 2026), a site’s technical AI readiness was a strong predictor of whether it was cited as a source — but said virtually nothing about whether the company was recommended. Recommendations also depend on your reputation out in the world: reviews, mentions, third-party validation. These two outcomes each require a different approach, and your test shows you where you stand on both. Read more about what determines whether AI mentions your company.

Why is your result a sample — and not a measurement?

Ask the same question tomorrow, and the answer might be different. AI models do not respond consistently from one instance to the next; the answer may depend on the wording, and the models are updated continuously. This doesn’t mean your test is worthless — it means you should interpret it for what it is: a snapshot. Conclusions such as ‘the AI doesn’t know me’ or ‘I’m well covered’ require you to see the same pattern repeat itself over weeks, for the same question, with the wording kept constant.

That’s why the notes from step 4 are important: without the date, model and wording, you cannot distinguish a real change from random variation. If you want to compare results over time, repeat the test regularly — for example, on the first Monday of the month — using exactly the same five questions.

What do you do with the result?

The test typically points in one of three directions:

  • You weren’t mentioned, and the results mostly cited portals and directories: Start with the basics — can the AI engines even read and understand your site? The free test measures this in 60 seconds. One specific place to start: in our scan of 465 search-visible Danish business websites (our own survey, July 2026), 89 per cent were missing an llms.txt — the generator will create one for you free of charge.
  • You were mentioned, but with errors: tidy up your information where the AI retrieves it from: your own website, CVR data, review profiles and structured data, so that your name, address and services are consistent across the board.
  • You were mentioned correctly: Congratulations — but don’t rely on it. Your competitors may overtake you, and the models change. This is about keeping an eye on things, not about correcting them.

Whatever the outcome, the next natural step is to turn the spot check into a regular monitoring process: the same questions, asked systematically, week after week. That’s exactly what Geoa does — and you can get started without paying a penny: create a free account, enter your five best questions, and they’ll be monitored automatically, so you can see the trends rather than guessing at them. In other words, your neighbourhood test isn’t a waste of time — it’s precisely the five questions your monitoring should be based on.

Frequently asked questions

Why do I get a different answer to the same question than my colleague?

Because AI responses aren’t deterministic: the same question can yield different answers from one instance to the next, and memory or previous conversations can influence the response for an individual account. That is precisely why a single response is just a snapshot — and why you should test in a new conversation and repeat the process over time before drawing any conclusions.

Should I test with AIs other than ChatGPT?

By all means — ChatGPT is the largest, but Gemini, Copilot and Perplexity draw on their own sources and may present a different picture. Perplexity is particularly useful in a manual test because it clearly displays its sources, so you can see exactly where the information comes from.

My business wasn’t mentioned — is that a problem?

It depends on the query and your business. If competitors in your town are mentioned in purchase-intent queries where you should be featured, it’s worth taking action. The first step is a basic check: can the AI engines read and understand your website? That’s the part you can fix yourself — recommendations also require reputation and publicity.

Can I get ChatGPT to mention me by telling it about my business?

No — not for anyone other than yourself. If you tell ChatGPT about your company, it may influence the responses in your own account via the context and memory of the conversation, but it won’t change the responses your customers receive. The real way forward is to make your information easy to find, read and quote — and to build up a reputation that the models come across.

Get your free GEO Score — 60 seconds, no credit card required.

Get started for free

Related articles

Stay up to date with more GEO insights for smaller businesses.

View all articles