Skip to content
ceeme

Prompt checker

This tool checks your prompt set for the three mistakes that make an AI visibility measurement worthless: a question that already rules out the competition, a question containing your own trade word, and a question asking for alternatives to what you are yourself. You see line by line whether it measures anything and why not, plus the engine weighting for the country you pick. No account and no email address.

Measure your own company → Rather talk it through first

In short

need question
What the person wants to achieve, in their own words, without knowing which kind of provider solves it. This is where the volume is.
exclusion
A question that rules out a kind of provider — "without a photographer". The competition is gone before the engine got to answer.
jargon lock
The ban on your own trade words in the need questions. A place name does not count: "in Utrecht" is what a person types.
query fan-out
The sub-questions an engine splits one question into before it answers.

Do your prompts actually measure anything?

Enter what you do, pick your customers' country, and paste your questions below. You get a verdict per line, with the reason. No account, no email, no trial.

Which three mistakes does the tool make visible?

These three, and all of them come from a real measurement that was worthless because of them.

The mistakeWhat it looks likeWhy it breaks the measurement
Exclusion"passport photo without a photographer"The competition is ruled out before the engine got to answer. You are measuring your own assumption. "Without an appointment" is fine — that is a circumstance, not a provider.
Your own trade word"best photo booth Utrecht"Anyone typing the trade word already knows what they want. You are the answer by definition, and you measured nothing. The money is with whoever types their problem.
Ruling yourself out"alternative to a butchery course with its own shop"Asked by a school that is itself a butchery course with its own shop. The answer could only be "absent", and that number went into the score.

In which engines does such a question get asked?

In the engines that matter for your market, not in all of them. For Belgium the weighting looks like this, with source and month — because a score whose weighting you cannot see cannot be checked.

Copilot is not listed, and that is not an oversight: at 6.96% it is the third player in Belgium, but there is no public endpoint that lets you ask what Copilot answers someone. What exists is Azure OpenAI, which is the underlying model — a detour we will not sell as a measurement.

EngineShare in BelgiumCan we measure it?
ChatGPT73,8%measurable, own API
Gemini10,4%measurable, own API
Perplexity4,5%measurable, own API
Claude4,2%measurable, own API

Source: Statcounter Global Stats, AI Chatbot Market Share Belgium, July 2026. Copilot (6.96%) and Phind (0.07%) are not listed — see the paragraph above.

How many of each kind belong in a set?

Forty per cent need questions, twenty per cent comparison questions from ten upward, the rest category questions. That is not a guideline but the reason a measurement is worth anything: a set made mostly of category questions mainly measures whether you know your own name.

Set ofNeed questionsCategory questionsComparison questions
9450
3012126
100404020
150606030

What does this tool not do?

What this tool doesWhat this tool does not do
Check your existing prompts against three rulesInvent new prompts for you — that is the scan's job, and it needs a model
Work without an account and without an email addressKeep your prompt set tied to a person, or put a form in front of you before showing anything
Say which question measures nothing, and whySay whether you appear in that question — that needs a real measurement in the engines
Use the same rules as the paid measurementGive a score. A number without a measurement is a guess with a decimal point

Frequently asked questions

What happens to my prompt set?

Your questions go to our server to be judged there, and we do not keep them tied to a person — we do not ask for an email address either. Doing the calculation server-side is a deliberate choice: the rules we check against are the rules of the paid measurement, and those should not be lying around as a downloadable file.

Are these the same rules as in the paid measurement?

Literally the same. The file in your browser is cut from the source, not retyped, and a test fails the moment the two diverge. That is deliberate: a free tool that says something different from the product is a promise you have to take back later.

Why is "without an appointment" fine and "without a photographer" not?

Because an appointment is a circumstance and a photographer is a provider. Ruling out what the person does not have or does not want to endure is a normal way to ask. Ruling out a kind of provider is not: you have already given the answer.

Does this work in French, English and German too?

Yes. The exclusion rule knows zonder, sans, without and ohne, with their articles. The French variant did once fail: in JavaScript, à is not a word character, so the expression never matched. That is noted in the code so nobody rebuilds the same bug. Language and country are separate here: language decides which exclusion words apply, country decides which place names count and which engines carry weight.

Why does a place name not count as a trade word?

Because people do type their city. It went wrong here once: the filter banned every five-letter word from the intake, including Bruges and Utrecht, so the good need questions were dropped and the nonsense stayed. The place list covers Belgium, the Netherlands, Germany, France and Luxembourg.

Read on

Free, no account and no card.