Prompt research is keyword research for AI search. Instead of mapping the short phrases people type into Google, you map the conversational questions they ask ChatGPT, Perplexity, Gemini and Google's AI Mode, including the context and constraints that decide which brands get recommended. It's the foundation for content planning, measurement and outreach in answer engine optimization.
Keywords vs prompts
| Keyword | Prompt | |
|---|---|---|
| Example | "project management software" | "What's the best project management tool for a 12-person design agency that bills clients by the hour?" |
| Length | 2–4 words | 10–40 words |
| Context | Stripped out | Included: team size, industry, budget, constraints |
| Intent | Inferred | Usually explicit |
| Follow-ups | Separate searches | Same conversation |
| Data | Search volume available | No platform-published volume |
That example prompt will surface different tools from the keyword, because "design agency" and "bills clients by the hour" point to time tracking and client billing features.
The five-step method
Step 1: Collect seed questions
Gather real buyer questions from sales calls, support tickets, onboarding surveys, communities and search data. Our guide to finding the questions buyers ask AI covers eight sources.
Step 2: Expand with modifiers
For each core question, add the modifiers buyers actually use:
- Segment: team size, company type, industry, role
- Constraint: budget, location, compliance, platform (iOS, Windows), integration
- Priority: easiest, cheapest, most secure, fastest to set up
- Comparison: vs a named competitor, alternatives to a named competitor
AI assistants can help brainstorm variants, but anchor them in real language.
Step 3: Group into intents
Cluster prompts that would be answered the same way. "Best CRM for a small real estate team" and "real estate CRM for a 5-agent brokerage" are one intent. "Real estate CRM with MLS integration" is a different intent because the constraint changes the answer.
Step 4: Prioritize
Without published prompt volumes, use three signals:
- Business value: how close the intent is to purchase and how valuable the segment is.
- Demand proxies: keyword volume for the closest search queries, frequency in your sales and support data, and community activity.
- Visibility gap: from your baseline, how often you're mentioned for this intent versus competitors. See AI visibility metrics.
High value, real demand and a large gap go to the top.
Step 5: Map intents to winners
For each priority intent, decide what should win it:
| Intent type | What should win it |
|---|---|
| "Best X for Y" | A use-case page, plus presence in roundups and reviews |
| "X vs Y" | Your comparison page, plus third-party comparisons |
| "X pricing" | A clear pricing page |
| "How do I…" | Documentation or a guide |
| "Is X worth it" | Reviews, case studies, community proof |
| "What is…" | An explainer page |
Then check what's actually cited today and close the gap. See comparison pages that win AI recommendations and how AI assistants decide which brands to recommend.
Don't throw away keyword research
When AI assistants search, they turn prompts into queries, often several narrower ones. Those queries look a lot like keywords. So:
- Keyword research still tells you what retrieval systems will search for.
- Ranking for the narrow, specific queries behind a prompt helps you get retrieved.
- Keyword volume is a useful proxy for prompt demand.
Prompt research adds the layer keyword research misses: context, constraints and the conversation.
A worked example
A company selling scheduling software for clinics might end up with intents like:
- Best scheduling software for a small physical therapy clinic
- Clinic scheduling software with online booking and reminders
- HIPAA-compliant appointment scheduling tools
- [Product] vs [Competitor] for multi-location clinics
- How much does clinic scheduling software cost per provider
- How to reduce no-shows with automated reminders
Each becomes a cluster of three to five prompt variants in the tracked set, mapped to a page that should win it and the third-party sources currently cited.
Keeping prompt research current
- Quarterly: review new sales and support themes, add or retire intents.
- On launches: add prompts for new products, features and markets.
- On competitor moves: add comparison prompts for new competitors.
- Keep a core set stable so trends remain comparable over time.
A prompt research worksheet
Keep one row per intent, with columns like these:
| Column | Example |
|---|---|
| Intent | Clinic scheduling with online booking and reminders |
| Representative prompts | "Best scheduling software for a physio clinic with online booking"; "appointment software that texts patients reminders" |
| Stage | Shortlist |
| Persona | Clinic manager |
| Business value (1–5) | 5 |
| Demand proxy | Related keyword volume, mentions in sales calls |
| Current mention rate | 20% (baseline) |
| Top cited sources | Two roundups, one review platform, a competitor's feature page |
| Page that should win | /features/online-booking |
| Next action | Rewrite feature page answer-first; pitch both roundups |
A worksheet like this turns prompt research into a backlog that content, SEO and PR can all work from.
The bottom line
Prompt research takes the discipline of keyword research and applies it to how people actually talk to AI: in full sentences, with context and constraints. Collect real questions, expand them with the modifiers buyers use, group them into intents, prioritize by value and gap, and map each to the content or source that should win it.
Wondering how many people actually ask a given prompt? See AI prompt volume for what can and can't be measured.
