Course lesson · Beginner
Prompt Research
Prompt research is finding the real questions your buyers ask AI assistants, in their own words, and turning them into a set of prompts you can test. Unlike keyword research, there's no public volume data for AI prompts, so the best evidence comes from conversations with customers: sales calls, support tickets, reviews and forums.
In this lesson
- How prompts differ from keywords
- Where to find real buyer questions
- Turning questions into testable prompts
- Grouping prompts by intent
- Exercise and quiz
How are prompts different from keywords?
| Search keywords | AI prompts | |
|---|---|---|
| Length | Often 2–5 words | Often a full sentence or more |
| Context | Little | Situation, constraints, preferences |
| Example (illustrative) | "crm nonprofits" | "We're a 12-person nonprofit using spreadsheets for donors; what CRM should we switch to on a small budget?" |
| Volume data | Available from search tools | Not published by AI providers |
Google has said AI Mode helps with longer, more complex questions (Google), and it uses query fan-out to explore subtopics (Google). Prompts reflect that complexity.
Where do real buyer questions come from?
- Sales calls and demos: what prospects ask before buying
- Support tickets: confusion and problems
- Reviews: what buyers compare and care about
- Forums and communities: Reddit, industry forums, Q&A sites
- People Also Ask and autocomplete in Google
- Your own team: customer-facing staff hear questions daily
Copy the wording exactly. Paraphrasing removes the context that makes prompts realistic.
How do you turn a question into a prompt?
- Keep the buyer's situation: role, size, constraint
- Remove anything identifying a real person or company
- Write it as you'd type it into an AI assistant
- Create a few variations for important questions, because wording affects answers
How do you group prompts?
| Intent | Example (illustrative) |
|---|---|
| Problem / learn | "Why do our invoices keep getting paid late?" |
| Category | "What tools help small agencies get paid faster?" |
| Comparison | "Is Tool A or Tool B better for a 10-person agency?" |
| Brand | "Is Tool A reliable?" |
A balanced prompt set covers all four, weighted toward the stages where buyers discover options.
What mistakes should you avoid?
- Writing prompts that favour you, such as "Why is Tool A the best?"
- Only branded prompts, which overstate visibility
- Ignoring personas: different buyers ask differently (ICP persona)
- Never updating: buyer questions change with the market
How do you estimate which prompts matter most?
Without volume data, use proxies:
| Proxy | What it suggests |
|---|---|
| How often the question appears in sales and support | Real demand from your buyers |
| Search volume for related keywords | General interest in the topic |
| Presence in People Also Ask or autocomplete | Common phrasing |
| Revenue linked to the use case | Business importance |
Score each prompt on two or three of these and prioritise the highest. Revisit the scores every quarter, because buyer questions change as markets and products change.
How do you keep the research organised?
Use one sheet with columns for the prompt, original wording, source, date, intent, persona and priority. This becomes the basis of your prompt set and makes it easy to explain where each prompt came from.
How do you phrase prompts like real buyers?
Write prompts the way people speak to assistants: full questions with context, such as "We're a 12-person agency on Google Workspace. Which CRM should we use?", rather than keyword fragments. Keep both a short and a long version of your most important prompts, because the answers often differ.
A quick example
Illustrative. A cybersecurity startup reviews 30 sales call notes and finds that prospects rarely ask "best endpoint security"; they ask "how do we meet our insurer's security requirements with a two-person IT team?" Its prompt set shifts accordingly, and the team discovers its visibility for these real questions is far lower than for generic category prompts.
Prompt research checklist
- [ ] Questions collected from at least three sources
- [ ] Exact buyer wording preserved
- [ ] Identifying details removed
- [ ] Prompts grouped by intent
- [ ] Mostly unbranded
- [ ] Variations for key questions
- [ ] Review date set
Exercise
Collect 20 real buyer questions from sales calls, Reddit or reviews.
- Gather 20 questions from at least three sources.
- Paste them into a sheet with source and date.
- Rewrite each as a prompt, keeping the buyer's situation.
- Tag each with an intent: problem, category, comparison or brand.
- Run five of them in an AI assistant and note whether your brand appears.
Quiz
- Do AI providers publish prompt volume data?
a) Yes, for every prompt b) No c) Only for brands Answer: b
- What's the best source of realistic prompts?
a) Customer conversations b) Your homepage c) Competitor ads Answer: a
- Why keep the buyer's exact wording?
a) It preserves context that affects answers b) It's shorter c) It's required by GA4 Answer: a
- Which prompt is biased?
a) "Why is Tool A the best?" b) "What tools help agencies get paid faster?" c) "Is Tool A or Tool B better for a 10-person agency?" Answer: a
- Most prompts in a set should be…
a) Branded b) Unbranded c) One word Answer: b
Next: Lesson 7: Brand positioning for AI
Sources
Questions
Is there keyword volume data for AI prompts?
Not in the way there is for search. AI providers don't publish prompt volumes, so research relies on customer evidence and proxies.
How many prompts should I collect?
Start with 20-50 real questions, then group them by intent. Quality and realism matter more than volume.
Should prompts include brand names?
Include some, but most prompts should be unbranded, because that's where buyers discover options.