You know your keywords. The sentences someone types into ChatGPT or Perplexity when looking for your service are probably unknown to you. Prompt mapping captures them, sorts them by buying stage and shows where you are missing from the AI answer.
Into a search box, people type two words. To a chat, the same person describes their situation: budget, industry, timeline, what has gone wrong so far. “seo agentur” turns into “we are an SME in eastern Switzerland, our website has brought in hardly any inquiries for a year and we are looking for someone to fix that”.
AI search answers these sentences, not your keyword. And it names a short selection of providers, not a list of results. If you are not among them, you do not exist in that answer.
The obvious reaction is to produce more content and hope it works. The better approach is to first find out which customer questions are actually being asked and for which of them you are missing today. That is what prompt mapping does.
Four sources that complement each other. None of them is enough on its own. Then the measurement starts.
For every keyword that matters to your business, we derive a prompt. “golfplatz bodensee” becomes “Which golf courses are there on Lake Constance?”. That way you measure AI visibility on the same topics that earn you money in classic search.
Search Console shows full-sentence questions your site already appears for. These sentences are not made up; they are demand you can measure. We group them into prompt clusters by intent. How we derive intent from Search Console data is explained in our guide to Reverse Search Intent Engineering.
What prospects ask in sales calls and what lands in support, we rephrase as prompts. These customer questions do not show up in any tool; they come from your team.
We check which questions currently trigger a mention of your competitors. This shows where there is a gap and where the spot is already taken.
The finished prompt sets run regularly against ChatGPT, Perplexity, Claude and Google AI Overviews. Answers change, so a one-off survey only shows the state on the day it was taken.
None of the prompts contains your brand name. If the prompt names the brand, you only learn whether the AI knows it when you tell it. A useful prompt describes the problem and leaves open whom the AI recommends.
A prompt set on its own tells you nothing yet. Only the classification shows where you are not mentioned and what a mention would be worth.
Each cluster gets a stage: first awareness of the problem, search for a solution, choice of provider, questions after the purchase. We weight mentions in the provider stage more heavily, because that is where the decision is made.
A conversation rarely ends after one question. We follow the next questions the AI itself suggests and check whether you are still in the answer two steps later.
The matrix becomes a ranking: which content is missing, which existing pages do not deliver the answer, where a mention outside your own website is needed. Then the implementation starts.
The mapping tells you where you are missing. It does not close the gap. That takes content an AI finds useful as a source, and mentions on sites you do not own.
How content is built so that it gets cited is covered on the page about citation engineering. If an AI system says something wrong about you, that falls under brand sentiment and reputation. The bigger picture is described on our GEO and LLMO management page. If you prefer to start on your own, you will find the steps in the Swiss GEO Framework (PDF in German) to download.
Prompt mapping is part of our GEO management, from CHF 1,800 / month, cancel monthly. 12 months: from CHF 1,710 / month (5% discount). We do not sell it separately, because an inventory only moves something once the work that follows begins.
A group of questions with the same intent but different wording. “Which GEO agency in Switzerland is good?” and “Who can help me with AI visibility?” belong to the same cluster. We measure per cluster, not per individual wording. Which wordings a cluster contains is defined before the measurement.
Because people write to an AI differently than to a search box. In a search box you type two words; in a chat you describe your situation with budget, industry and timeline. You need both sources: keywords show the topic, full-sentence queries show the language.
Because otherwise you measure whether the AI knows a brand when you say it first. The real question is whether it names you on its own. A useful prompt describes the problem or the category and leaves open whom the AI recommends.
The guideline is ten active money keywords per client, from which the prompts are derived. On top come the clusters from Search Console, sales conversations and competitor analysis. The exact number depends on what you offer, not on a package.
We look at the sentences your customers use with AI systems and where you do not appear in the answer. Free and without obligation.