Let’s talk

GEO · · 4 min read

How to check what AI assistants say about your brand (and what you can't measure)

How to check what ChatGPT, Gemini or Perplexity say about your brand: queries, mentions, citations, accuracy, sources and how often to review.

A magnifying glass over a row of speech bubbles, with a document sheet linked to one of them by an arrow.

There is no “ranking” for your brand in ChatGPT, Gemini or Perplexity that you can check every Monday. What you can do is define a sample of queries, repeat it with a fixed method and record what each assistant says, where it gets it from and whether it is correct. It is an imperfect measurement, but a useful one if you know where its limits lie.

Can you measure presence in AI the way you measure position in Google?

Not in the same way. In a traditional search engine, a keyword has an ordered list of results and you can track your position. In an AI assistant, the answer is generated each time. It changes depending on how you phrase the question, the engine, the model, the language, the location and, in some cases, the conversation history.

This does not mean nothing can be measured. It means that what we measure are trends across a sample, not positions. A single screenshot where you appear (or don’t) proves nothing. Twenty queries repeated with the same criteria over several months do start to tell you something.

Which queries should you test?

The most common mistake is to ask only about the brand name. That measures whether the assistant knows you, but not whether it recommends you when someone doesn’t yet know you exist. It is best to build a sample with three groups:

  • Brand queries: “What is [brand]?”, “[brand] reviews”, “[brand] vs [competitor]”. They help you check what is being said about you.
  • Category queries: “best [service] in [city]”, “how to choose [product]”. Here you see whether you take part in the conversation in your sector.
  • Problem queries: what a customer who doesn’t yet know which solution they need would type. They are the hardest and the most valuable.

A sample of between 20 and 40 queries, worded the way a real customer would say them and not the way your marketing team would, is usually enough to get started. Keep them in a closed list. If you change the questions every month, you won’t be able to compare.

What should you record for each answer?

Every time you run a query, always note the same fields, with the date and the engine:

  1. Mention: whether your brand appears, where in the answer and in what context (recommended, cited as an alternative, compared).
  2. Citation: whether the assistant links to or references one of your pages as a source, and which one.
  3. Accuracy: whether what it says about you is correct (services, location, prices, opening hours, contact details) or out of date or wrong.
  4. Sources: which other websites it uses to talk about your sector. Often they are not yours, but directories, media outlets or forums.
  5. Competitors: who appears when you don’t.

Accuracy is usually the most urgent. An AI saying something false about your company is a problem with a traceable cause: an old listing, a contradictory page, a third-party source. It is what we review first in our GEO service, because fixing it depends on the quality and consistency of the information you publish.

What can’t really be measured?

It is worth being honest about what no report can give you:

  • A real volume of queries. Assistants don’t publish how many people ask what, so there is no reliable equivalent to search volume.
  • A stable position. What you see today may not be repeated tomorrow with the same question.
  • The exact cause of a mention. You can observe which sources appear, but you cannot know the internal workings of the model.
  • The direct effect on sales. Part of the traffic arrives without a clear referrer. You can cross-check the sample against visits and enquiries in your analytics, but it will be a clue, not proof.

There is no separate trick for getting into the answers either. Google’s own documentation on its AI features in Search says they rely on the usual fundamentals: useful content and pages that can be crawled and indexed (you can read it in its official guide). That is why measuring is useful for spotting gaps and errors, not for promising results.

How often should you review it?

A monthly review of the full sample is reasonable for most SMEs. Answers do move, but not so much as to justify checking them daily, and a weekly rhythm only adds noise. Add an extra review when:

  • You change something important on your website: services, prices, location or structure.
  • You spot an incorrect detail and have corrected it at the source; that way you check whether it gets updated.
  • An engine launches a significant change in how it answers or cites sources.

Always keep the history, with date and engine. The value lies in comparing month by month, not in an isolated snapshot.

Where to start this week

Choose ten real customer queries, run them through the assistants you use most and note the mention, citation, accuracy and sources. That already gives you a first reading. If you would rather we did it for you, request the free GEO audit: we review a sample of queries from your sector and send back what is said about your brand, with sources, dates and engines. If you also want to connect that data with your analytics, we work on it through analytics and market research.

Frequently asked questions

Can I find out how many people ask about my brand in ChatGPT?

Not reliably. Assistants don't publish query volume by topic. What you can measure is what they answer to a fixed sample of questions and how that evolves over time.

Is it enough to ask ChatGPT about my company once?

No. The answer can vary between attempts and between engines. A single query is an anecdote; a sample repeated with the same method lets you see trends.

What should I do if an assistant says something false about my company?

First, find where the information may come from: your website, an old listing, a directory or a media outlet. Correct it at that source, keep your information consistent everywhere and check the query again in later reviews.

How often should I review my presence in AI?

For most SMEs, once a month with a closed sample of queries. Add an extra check if you change something important on your website or correct an error you have detected.

Related service

Does your company show up when your customers ask AI?

More and more people ask AI assistants before they choose. We review what they answer about your sector, which sources they cite, and what is missing from your website and digital presence for you to be well represented.

Explore the GEO service

Tell us what you want to achieve.

Someone from the team replies with a first proposal for the approach.

Let’s talk about your project