How to conduct prompt research to discover what your clients ask AI?

Why doesn’t prompt research work like keyword research?

No public volumes for questions asked to AI

SEO tools display estimated volumes for each keyword. There is nothing like that for questions posed to ChatGPT, Claude, or Perplexity: the providers keep this data to themselves. There is no way to know how many people asked this month which expense management software to choose for a 50-employee SME. You have to start from your clients’ real intentions and contexts, using scattered sources that require some reconstruction work.

Longer, more specific questions

A prompt is not a keyword. AI users often formulate complete questions, with context, constraints, and sometimes their profile. Where a web user might type “expense management software,” an AI user might ask: “I run a 50-employee SME, which simple expense management software compatible with our accounting system do you recommend?” The prompt already contains the intention, the need, and the selection criteria.

Where to find ideas for your prompt research?

No single source directly provides the prompts being used, but several reveal intentions, phrasing, and expressed needs:

• Long-tail SEO keywords: multi-word queries already contain a specific intention.
• Google’s “People Also Ask”: these reflect frequent questions and can easily be reformulated as prompts.
• Customer service and sales teams: they hear the real questions, hesitations, and selection criteria every day.
• Customer reviews: they show what matters to users and the problems they are trying to solve.
• Industry forums: Reddit, Quora, and specialized groups are full of detailed questions.
• Follow-up questions suggested by AIs: after an initial answer, ChatGPT or Perplexity suggests complementary questions that reflect common user journeys.

Example: you offer expense management software. Starting from the keyword “expense management software,” you can construct prompts based on intent:

• Discover: “How can I simplify expense management in an SME?”
• Compare: “What are the differences between the main expense management software options for an SME?”
• Choose: “Which expense management software should I choose for 50 employees with integration to our accounting system?”

Each intent corresponds to a stage in the customer journey, and AIs can answer at all of these stages.

How to select priority prompts for your GEO strategy?

Once you have collected the questions, you need to sort them. Here is a five-step method to build a manageable corpus:

1. Group questions by intent: discovery, comparison, choice, and problem-solving.
2. Keep those related to your offer and the purchase decision: questions that are too general or far from your business can wait.
3. Vary the phrasing: the same intent can be expressed in multiple ways, long or short. Keep those that really sound like your clients.
4. Cover your different client profiles: a manager, a buyer, and an end user do not ask the same questions. The Personae feature lets you simulate these profiles.
5. Limit the corpus to a manageable size: a few dozen priority prompts are enough to get started.

How to evolve your prompt research over time?

Questions evolve with trends, new products, and the market. Review your corpus with each launch, offer change, or new seasonality, and integrate emerging questions reported by your teams. However, keep a stable core of prompts so you can compare results from one period to another.

Once you have identified the priority prompts, you need to check over time whether your brand appears in the answers. This is the role of Search AI’s prompt tracking feature.