How to Do Prompt Research: Find the Questions Buyers Ask AI

Prompt research is the process of finding the real questions your buyers ask AI assistants, using sources like sales calls, support tickets, Search Console queries and community threads, then grouping and prioritising them for content and tracking. AI platforms don't publish category-level prompt data, so good research triangulates from what customers actually say. This tutorial walks through the process in five steps.

Key takeaways

• Start with customers' own words, not keyword tools.
• Prompts include context: role, budget, constraints, location.
• Cluster by intent and funnel stage, then prioritise.
• Feed results into content plans and your tracked prompt set.

How are prompts different from keywords?

Prompts are longer, include constraints and often ask for a recommendation or a plan. That context changes which brands are suggested.

Step 1: Which sources should you mine?

Collect questions verbatim in a sheet with a column for source.

Step 2: How do you turn raw material into prompts?

Rewrite each finding as a natural question a buyer might type, keeping constraints:

1. Start from the verbatim phrasing.
2. Add the persona's typical context ("for a 20-person team").
3. Keep it neutral; don't name your brand in discovery prompts.
4. Write two or three variants for important topics, since wording affects answers.

Illustrative: a support ticket says "Does it sync with Xero? We keep double-entering." That becomes "Which invoicing tools sync automatically with Xero so I don't have to enter invoices twice?"

Step 3: How do you cluster and tag?

Tag every prompt with:

• Topic: the product area or problem.
• Stage: awareness, consideration or decision.
• Type: discovery, problem, comparison, branded, feature.
• Persona: who's asking.

Then group near-duplicates. You'll usually find a handful of themes covering most questions.

Step 4: How do you prioritise?

Score each cluster on three factors (1–3 each):

Run the top prompts in the AI assistants your buyers use and note who's recommended. Clusters with high value, strong fit and a visible gap come first.

Step 5: What do you do with the results?

1. Content plan: each priority cluster gets a page, or an improvement to an existing page, that answers the questions directly.
2. Tracked prompt set: select a balanced, sized subset for ongoing measurement (building a prompt set).
3. Sales and product feedback: share recurring questions with sales and product teams; they often reveal positioning gaps.

Worked example

Illustrative. A time-tracking startup collects 180 candidate questions: 70 from sales notes, 50 from support, 40 from Search Console and 20 from Reddit. Clustering yields nine themes. "Time tracking for agencies that bill clients by project" scores highest: high value, strong fit and, in test runs, the brand is absent from ChatGPT answers. The team writes a dedicated guide with a pricing comparison and client-billing workflow, and adds 25 prompts from that cluster to its tracked set.

How do you run a quick internal survey?

Your colleagues are a fast, free source, especially in sales, customer success and solutions roles. Send three questions:

1. "When a prospect first contacts us, what's the question they're usually trying to answer?"
2. "Which competitors do prospects mention, and why?"
3. "If you were the buyer, what would you ask ChatGPT?"

Give people a week and a shared doc. Answers from customer-facing staff are often more specific than anything a keyword tool returns, because they reflect real deals. Credit contributors when the research turns into content; it keeps them sending questions.

How do you check whether a prompt reflects real demand?

You can't see AI platform volumes, so look for corroboration: the same question appearing in two or more sources (for example sales notes and Search Console) is a good sign. Single-source questions can stay in the explore group until they show up again.

How often should you repeat research?

Refresh quarterly. New features, competitors and market events change what buyers ask. A light refresh, with new sales questions plus a Search Console review, takes a couple of hours; a full rerun once a year keeps the set honest.

Common mistakes

• Starting from keyword tools and adding "best" in front.
• Dropping the constraints that make prompts realistic.
• Including your brand name in discovery prompts.
• Researching once and never refreshing.
• Prioritising questions where your product isn't a genuine fit.

Checklist

• [ ] At least four sources mined, verbatim
• [ ] Questions rewritten as natural, neutral prompts with context
• [ ] Tagged by topic, stage, type and persona
• [ ] Clusters scored on value, fit and gap
• [ ] Priority clusters mapped to content and the tracked set
• [ ] Quarterly refresh scheduled

Where to go next

• Building a prompt set: how many prompts do you need?
• Recommendation rate: the metric that matters

Sources

• Google Search Console Help, Performance report