Does ChatGPT mention your business? How to measure what AI says about you

When I was building my own website, I asked an AI assistant to look up my Instagram account so I could put the link in the footer. Within a minute it had found one. The name looked strikingly like mine and it presented it without a shred of doubt. Only the account wasn't mine.

It hadn't made anything up. It had found something that was almost right, and for a language model, almost is often good enough. That time it was about a link in a footer. But suppose a potential client asks ChatGPT who I am, what I do and whether I'm any good at it. What do they get told, and how do you actually find that out as a business owner?

The short answer: you measure it yourself, by asking a fixed set of questions in ChatGPT, Gemini, Perplexity and Google, recording the answers every month and looking at which sources they use. There's no official dashboard for ChatGPT, so you treat it as a sample and look for patterns.

The AI didn't make a mistake, the internet was unclear

The easiest thing is to dismiss such a wrong answer as an AI talking nonsense. I think that's too simplistic. A language model tells people about you based on what it comes across about you online, and fills the gaps with whatever seems most likely. If there's someone with a name similar to yours, your profiles don't match or there's little clear information about you online, it starts guessing.

That's the core of GEO, generative engine optimisation. It's not about outsmarting an AI tool. It's about making sure there is so much clear and accurate information about you that guessing is no longer necessary. I described how that works earlier in Why ChatGPT doesn't know your business. This article is about the step before that: knowing where you stand.

Why you can't just measure this in one go

For Google you have Search Console. There you can see which searches you appear for and how often people click. According to Google, impressions in AI Overviews and AI Mode count towards the performance report in Search Console, under the Web search type. In June 2026 Google also announced separate reporting on generative AI in Search Console.

For ChatGPT, Gemini as a standalone app and Perplexity, nothing like that exists. On top of that, a language model doesn't give the same answer every time. The outcome depends on how the question is phrased, what someone asked earlier in the conversation and which sources are found at that moment. That's why you don't measure one answer, but a series.

Start with the questions your customer actually asks

Write down ten to twenty questions. Don't think in keywords, because people type full sentences to a chatbot, often with their whole situation included. Spread them over three types:

• Category questions. Someone is looking for a provider: "Which copywriter in Maastricht helps with SEO?"

• Problem questions. Someone has a problem and doesn't yet know you're the solution: "How do I get ChatGPT to mention my business?"

• Brand questions. Someone already knows your name: "What does Nicci Severens do?"

  
  

Those brand questions feel a bit like googling yourself on a lost Sunday afternoon, but start with them anyway. If a model already talks nonsense when someone mentions your name, there's no point hoping for a mention with the rest.

Ask them everywhere, and make sure the tool doesn't already know you

Use at least ChatGPT, Gemini and Perplexity, and also look at the AI Overview at the top of Google. They work differently and don't get their information from exactly the same place.

Make sure your own history doesn't count. Use a temporary chat or an account without saved memory. Otherwise the model mainly tells you what it already knows about you, which is cosy, but not a measurement.

Record more than just yes or no

Make a simple spreadsheet with, per question and per tool:

• whether you are mentioned;

• in which position, first or fifth in a list;

• whether what is said about you is correct;

• which sources are listed (Perplexity and ChatGPT with search often show links);

• which competitors do get mentioned.

  
  

Those last two columns are the most valuable. The sources show where a model gets its knowledge from. The competitors show who has the story that you're still missing.

Look at your own figures too

If someone clicks through from a chatbot to your site, that can show up in Google Analytics as a referral from, for example, chatgpt.com or perplexity.ai. Not every tool passes on that information neatly, so see it as a lower limit.

Don't expect large numbers. According to Google, people who click through from an AI Overview do stay on a site longer. That fits what you'd expect: anyone who arrives via a chatbot has often already thought things through quite a bit.

One measurement says nothing, three do

Ask the same questions again every month, in the same tools, and add to your spreadsheet. After three or four rounds you'll see whether you're being mentioned more often, whether the information is correct and which sources keep coming back. Only then do you know where to start.

What to do with what you find

You're not mentioned. Content that answers the question is missing, or it's there but not clear enough. Write a page or article that answers exactly that question, with the answer in the first sentences.

Something wrong is being said about you. Find out where it comes from. Often it's an outdated profile, an old listing or confusion with someone with a similar name. Make sure your website, LinkedIn, Google Business Profile and other profiles say the same thing, and use schema markup to record which accounts really are yours.

Competitors do get mentioned. Look at which sources the model uses for them. Are they in lists, trade media or on review sites where you're missing? That's where your next step lies.

What measuring doesn't give you

Measuring tells you where you stand, not how to force a mention. Nobody can guarantee that an AI tool will mention you, and anyone who promises that is selling something other than they think. Google itself says that no special optimisation is needed to appear in its AI features. The basics stay the same: a findable site, clear copy and information that's correct.

Unlearning is harder than learning

That Instagram account was a small mishap with a simple solution. But it did show how it works. An AI tool that knows nothing about you can still be taught something. One that knows the wrong thing about you has to unlearn it first, and that starts with knowing what it thinks now.

Frequently asked questions

Can I measure whether ChatGPT mentions my business? 

Yes, by asking a fixed set of questions yourself in ChatGPT, Gemini and Perplexity and keeping track of the answers every month. There's no official dashboard like Search Console for that.

How often should I measure?

 Once a month is enough to see a trend. Measuring more often mainly produces noise, because answers differ each time.

Why does ChatGPT give a different answer every time?

A language model formulates every answer anew, depending on the question, the context and the sources it finds at that moment. That's why you look at patterns over several rounds.