How to read your AI visibility score

Why it matters

Your free report, "What AI Says About Your Business", sums up dozens of AI answers in one number from 0 to 100. That number is useful for one thing: seeing where you lose points, so you know which fix to do first. It is not a grade on your business, and one answer inside the report is not a verdict either.

This guide explains what each part of the score measures, what "recommended", "mentioned" and "absent" mean, how to read the competitor share, and how to check again yourself once your fixes are live.

Illustration with a fictional business.

Reference

What the free report checks

This is our own method. Every free report checks the same things:

• Three AI assistants: ChatGPT, Gemini and Google AI Overviews. In September 2026, ChatGPT sent about 80% of the visits that AI chatbots refer to websites in Luxembourg, and Google's Gemini assistant about 9% (Statcounter). Google AI Overviews are the AI answers at the top of Google search results. The free report does not check Claude, Perplexity or Microsoft Copilot.
• 10 customer questions in English, French and German: 6 in the language of your report and 2 in each of the other two. We write them for your business, in your customers' words, and they do not contain your name. A question such as "fiduciary for a new SARL in Luxembourg" shows whether AI suggests you to people who do not know you yet. The questions are listed in your report, in the table "Every question we asked".
• 30 answers. Each question is asked once on each of the three assistants, on the test date shown on the cover. A person on our team reads every answer.
• Your facts, asked by name. We also ask the assistants about your business by name and check six facts: name, address, opening hours, main services, prices, and phone and email.
• Only public information. Your website, public listings and the answers the assistants gave. We do not need access to any of your systems.

The score

Three parts, 100 points

On our home page the three parts are called Found by AI, Facts right and Against competitors. Inside the report, the pages are headed Discovery, Brand accuracy and Competition. They are the same three parts.

The total puts you in one of four bands:

Each answer

Recommended, mentioned or absent

Every one of the 30 answers gets one of three results:

Illustration with a fictional business.

Then look for patterns, not single cells. The visibility page splits your results by assistant and by language, so you can see, for example, that ChatGPT names you in English but no assistant does in French. A pattern points to a fix. One "absent" on its own does not.

Why answers change

One answer is not a verdict

Ask ChatGPT or Gemini the same question twice and you can get two different answers. Google says so about Gemini: "an LLM's response to the same prompt will not necessarily be the same every time" (Google: What is Gemini and how it works). Answers also change over weeks, as the assistants are updated and the websites they read change.

That is why your report shows what the assistants said on the test date, and why we read the totals. "Named in 6 of 30 answers" tells you much more than any one answer. If you ask one of the questions yourself today and get a different result, that is normal. It does not mean the report is wrong, or that the problem has gone away.

Competitors

What "share of all mentions" means

We count every time a business was named across the 30 answers. Your share is your count divided by the total for all businesses. It is also called share of voice. The report shows your top three competitors.

Illustration with fictional businesses.

Do not confuse the share with the number of answers. In this example Veldrin appears in 6 of the 30 answers, which is 20% of answers, and holds 21% of all mentions. The two numbers measure different things: one counts answers, the other counts names, and one answer can name several businesses. The competitors page also shows, per assistant, the gap between your share of answers and the top competitor's.

The share tells you who AI treats as the default choice for your questions. It does not tell you why. The finding under the chart gives our reading: what those competitors have that you lack, such as reviews, listings or a page that answers the question.

What to do first

Follow the order of your top fixes

The top fixes page lists five fixes, most important first. Each has a timing (this week, within 30 days, within 90 days) and roughly how long it takes. Each problem in the report has a short label, such as V1 for a visibility problem or A1 for a wrong fact, and the same label appears next to the fix that solves it.

Do the fixes in that order. Wrong facts and a website AI cannot read usually come first, because they affect every answer. Each fix below has a free guide that shows you how:

Fixes make your information available and clearer to the assistants. No one can promise that ChatGPT or Gemini will then recommend you, which is why you check again.

Steps

Read your report in 10 minutes

1. Note your score, band and test date

   They are on the cover. Write them down with the date: they are your starting point.

2. Find the part that loses the most points

   Subtract each part from its maximum: 50, 30 and 20. The biggest gap is your priority.

3. Look for a pattern in the 30 answers

   On the visibility page, check whether you are missing on one assistant, in one language or on one kind of question. Ignore single results.

4. Read the wrong facts

   On the accuracy page, the table of wrong facts says what AI says, what is true and where to fix it. These are usually the quickest wins.

5. Note your share and your top competitor

   Write down your share of all mentions and the top competitor's. Then read what they have that you lack.

6. Start fix 1, using the table above

   Open the matching guide and put the fix in your calendar for the week shown in the report.

Check again

Re-check it yourself

Once your fixes are live, wait one to two weeks, then run Guide 01 with the same 10 questions from your report. Keep everything else the same too: ChatGPT, Gemini and Google AI Overviews, the same languages, a new chat for every question. That way you compare like with like. After that, repeat it every month: AI answers change month to month, and a monthly check shows whether your fixes held.

Compare totals with totals: how many of the 30 answers named you, how many recommended you, and how many of the facts each assistant got right.

Common mistakes

What usually goes wrong

• Treating one answer as the result. Answers vary between runs and days. Read the totals.
• Re-checking with different questions. New questions, another language or another assistant give a number you cannot compare with your report.
• Counting a mention as a recommendation. Being named in passing or in the sources is not the same as being suggested.
• Starting with competitors. The competitor part is worth 20 points. Wrong facts and a website AI cannot read usually cost more, and are quicker to fix.
• Checking the day after a fix. Assistants and search engines need time to pick up changes. Wait one to two weeks after the change is live.