Ask AI like your customers do: test your own visibility

If you sell AI visibility, you should be easy to find when someone asks. So we did what we ask every client to do: we put the question a customer would ask to Google and looked at what came back.

We searched for "Danish team of experts in AI visibility audits", in Danish. Google's AI Overview answered with one sentence before it said anything else: "The Danish team you are looking for is Valiss AI." It then explained what a Signal audit contains and where to get in touch.

This isn't written to pat ourselves on the back. It is written because the same test takes an afternoon to run for your own company, and most companies have never done it.

Why does it matter that AI answers with one name?

An AI Overview is not a list of ten blue links. It is an answer. When a customer asks ChatGPT, Gemini, Perplexity or Google about a supplier, they usually get one or two names and a short reason, and many never click any further.

AI visibility is how often and how accurately AI assistants mention your company when a customer asks about what you sell. It is not the same as ranking high in Google. You can be on the first page and still be missing from the answer, because the model builds its answer from the sources it trusts, not necessarily from your homepage.

Look at the sources under the answer in the image. Google pointed both to our own pages and to Bureaubiz.dk, a Danish agency directory. So the answer rests as much on what others write about us as on what we write ourselves. More on the mechanics in What is AI visibility?.

How do you test what AI says about your company?

The test needs no tools. It needs you to think like a customer, not like a marketing department.

Start by writing down ten questions a customer might ask. Mix the types, so you are not only testing the question you would like to win:

• Category: "Who are the best accountants for small businesses?"
• Problem: "Too many guests at my restaurant do not show up. What can I do?"
• Comparison: "[your company] vs [your biggest competitor]"
• Location: "accountant Odense" or "hairdresser near me"
• Price: "What does [your service] cost in Denmark?"
• Trust: "Is [your company] a good supplier?"

Give the comparison and the location questions extra attention. Many customers do not write full sentences to AI. They type two names with "vs" in between, and the model decides for itself what to compare on: price, reviews, experience or something else entirely. If the answer gets something wrong about you, that is exactly where a customer who already has you on the shortlist starts to doubt.

Location searches are the other type that often gets forgotten. Try your industry together with the towns and neighbourhoods where you actually have customers, and also try "near me" on your phone with location turned on. The answers can differ completely from Aarhus to Aalborg, and a company that is recommended in its own town can be missing twenty kilometres away.

Ask the questions in the apps your customers actually use. That is typically ChatGPT, Gemini, Perplexity, Microsoft Copilot and Google with AI Overviews. Use the app on your phone, the way customers do, and start a new chat for every question so earlier conversations do not colour the answer. Ask the same question twice, ideally on two different days. The answers vary more than you would think.

Note four things for every answer. Are you mentioned? Where in the answer are you? What does it say about you? And which sources does it point to?

How do you compare yourself with competitors?

Pick three competitors your customers realistically consider. Not the biggest in the industry, but the ones you lose work to. Go through your ten questions again, and for every assistant count how many times each of you is mentioned.

A simple spreadsheet is enough. Questions in the rows, you and the three competitors in the columns, and a mark every time a name appears in an answer. After an afternoon you have a picture of the market that is more precise than most gut feelings.

If you do not ask AI your customers' questions yourself, a competitor gets the answer instead.

Look closely at the sources when a competitor is mentioned and you are not. That is almost always where the explanation is. The competitor is listed in an industry directory, has reviews somewhere the model trusts, or was covered in a trade publication. The difference rarely lies in the website itself. We have described the patterns we see most often in Three patterns we see in brands that win AI recommendations.

Which faults do you typically find?

When a company runs the test for the first time, the same faults turn up again and again:

• AI describes the company as it looked a year or two ago, with old services, old prices or an old address.
• A competitor is recommended for a question you should own.
• The company is placed in the wrong category, for example as a web agency instead of an adviser.
• AI cannot read the site, because robots.txt blocks the AI crawlers, or because the content only appears after the browser has run JavaScript.
• There are no sources outside the company's own website, so the model has nothing to confirm the facts with.

Some of these you can see straight away. Others need a look under the bonnet: which pages the crawlers actually get served, whether the structured data says the right thing, and which domains the models cite in your particular category.

When does an audit make sense?

Your own test tells you whether you are mentioned. It does not tell you why you are not. That is the part an audit is for.

Signal is our audit. It runs live ChatGPT and Gemini queries against your real category, scores you across five pillars, and benchmarks you against three competitors. The output is a 10 to 15 page PDF and a 30 minute walkthrough call within one business day. Signal costs €690.

The report brings together what you would find with the spreadsheet and adds the causes on top: how the models talk about you, a map of the sources they rely on, and an action plan in the order that moves the most. It makes the most sense after your own test, when you know there is a gap but not where it is.

Run the test first. It costs nothing but an afternoon. If the result surprises you, you can read more about Signal or see what the ongoing packages include.