The question of whether your business appears in ChatGPT can be answered in half an hour. You don't need any tools for it, only a fixed way of working. That way of working determines whether your measurement is worth anything, and that's where most self-tests go wrong.

The full step-by-step plan, followed by an explanation for each step:

1. Formulate five questions in your customer's words, from different phases of the decision.
2. Put each question to ChatGPT with the search function on, to Perplexity and to Google, each time in a new session.
3. Note for each question whether you appear, which businesses do appear, and which pages the sources link to.
4. Put the results in a scorecard: the questions as rows, the platforms as columns.
5. Repeat the measurement monthly with the same questions and the same method, and keep every version.

01 / preparationFormulate five questions that your customer actually asks

Don't start with your business name but with your customer's question. Whoever types in their own name is only testing whether ChatGPT knows them. The measurement that matters is whether you appear when someone who doesn't yet know you presents a problem that you are the answer to.

Write down five complete questions, in the customer's words. Not "accountant Eindhoven" but "which accountant in Eindhoven is good for a sole trader". The questions customers ask you on the phone are usually the best source. Record the five questions in a document, because you'll be using them again.

Choose questions from different phases of the decision. An orientation question ("what does an accountant cost for a sole trader"), a comparison question ("what should I pay attention to when choosing an accountant") and a recommendation question ("which accountant would you recommend in Eindhoven") each produce a different kind of answer, and therefore a different picture of where you do and don't appear.

02 / the measurementAsk the questions and note who appears

Put each question to three places: ChatGPT with the search function on, Perplexity, and Google. On Google, watch for the AI answer block above the results, if it appears.

Use a new, empty session for each question, without logged-in history where possible. An assistant that knows your previous conversations colours the answer with what it already knows about you, and that distorts the measurement.

Note three things for each question: whether your business is in the answer or in the sources, which businesses do appear, and which pages the source citations link to. That last column especially is instructive. You see not only who appears, but also with what kind of page: a knowledge article, a comparison, a review page.

The names of your competitors in the answer are more valuable than your own score: they show what does get cited.

03 / the pitfallWhy a single isolated test tells you nothing

An AI answer is not a fixed outcome. The same question produces a different answer in a new session, sometimes with different sources. Whoever tests once and sees themselves appear may have disappeared a week later; whoever doesn't see themselves may well appear in the next session.

That's why a single snapshot doesn't count as a measurement. What does count: the same questions, the same platforms, the same way of recording, at a fixed moment. If you change the questions along the way, you're comparing two different measurements with each other, and the difference proves nothing. Also be cautious about following up with more questions in the same session: the conversation steers the answers, and a steered measurement isn't a measurement.

04 / repeatingFrom snapshot to measurement series

Put the results in a simple table: five questions as rows, three platforms as columns, a tick where you appear. Repeat the measurement every month in the same way and keep every version. After three months you have a series, and a series shows whether there's movement in your visibility, which a single test never can.

This is what that scorecard looks like, filled in fictitiously for illustration:

Also note the date and the precise way of working for each measurement: which platform, search function on or off, new session or not. That may seem excessive for a table with fifteen ticks, but it's the difference between an impression and a measurement. Whoever wants to be able to say in four months that something has changed must be able to show that it was measured in the same way.

If you want to take this further, first read what AI visibility is for the broader picture, and then how you measure AI visibility for the step from self-test to a fixed benchmark. If you see competitors appearing on every question where you're missing, why ChatGPT mentions your competitor and not you covers what that's usually down to.

This self-test has deliberately been kept simple, and is enough for a first impression. Whoever measures five questions and doesn't encounter themselves anywhere knows what they need to know, and doesn't need to hire anyone for that.

This text was produced with AI assistance and checked and approved by a human before publication.