Persona stacking and query fan-outs: a better way to build a prompt panel
Most prompt lists come from keyword habits. Build yours from personas and the sub-queries AI engines actually run. A step-by-step method with a worksheet and tool support.
Quick verdict
A prompt panel built from a keyword export measures the questions your SEO team already thinks about. A panel built from personas and fan-outs measures the questions buyers ask and the sub-questions the engines ask on their behalf. It takes an afternoon and improves every metric downstream. Promptwatch supports both ideas natively: personas with persona stacking, and query fan-outs shown per model.
Two terms
Query fan-out. When an AI answers a prompt, it often runs several sub-searches behind the scenes. A question about the best accounting tool for a small agency may trigger searches on pricing, integrations, and reviews. Those sub-queries are the fan-out. Pages that answer them are candidates for citation, even when the original prompt never named them.
Persona stacking. A persona is a description of a buyer: role, context, constraints. Stacking layers several personas onto the same base prompt so you can see how answers change. "Best CRM" asked as a startup founder and as an operations lead at a large firm can return different brands.
Step 1: write the personas
Keep it to three to five. For each, write:
- Role and seniority.
- Company size or situation.
- The job they want done.
- One constraint, such as budget, compliance, or team skill.
Write them from real conversations with customers, not from imagination. If you have no data, interview two customers per persona.
Step 2: generate seed prompts per persona
For each persona, write five questions in plain language:
| Type | Example shape |
|---|---|
| Category | "What is the best [category] for [situation]?" |
| Comparison | "[Option A] vs [Option B] for [constraint]" |
| Problem | "How do I [job] without [pain]?" |
| Evaluation | "What should I check before buying [category]?" |
| Brand | "Is [brand] good for [situation]?" |
That gives you 15 to 25 prompts. Tag them by type and persona. Promptwatch supports topics, tags, and prompt types such as organic, brand, and competitor comparison, along with intent.
Step 3: add suggestions from data
Your own data holds prompts you would not invent. Promptwatch's Prompt Explorer suggests prompts from Search Console, ranking keywords, People Also Ask, or any page you point it at. Prompt Volumes add search volumes, difficulty scores, and citation rates across AI search, Google, and Bing.
Step 4: read the fan-outs
Once prompts are tracked, open the fan-outs. Promptwatch lists query fan-outs per model. Look for:
- Sub-queries that repeat across prompts.
- Sub-queries where you have no page.
- Sub-queries that name competitors.
Each of those is a page idea, often smaller and more specific than anything on your keyword list. GEO Inspector, a free Chrome extension from Promptwatch, shows ChatGPT fan-outs and the difference between fetched and cited sources, and runs locally. It is useful for a quick look before you commit to a plan.
Step 5: stack personas and compare
Run the same base prompt with each persona. Compare which brands appear and in what position. A table makes patterns visible.
| Persona | Mentioned? | Position | Competitors named | Sources cited |
|---|---|---|---|---|
| Founder at a small startup | ||||
| Operations lead at a mid-size firm | ||||
| Procurement at a large company |
If you are strong for one persona and absent for another, you have a targeting gap. Your pages may speak only to one buyer. A page that addresses the missing persona directly is the fix.
Step 6: audit the panel quarterly
Prompts decay. Products change. Review the panel every quarter, retire stale prompts, and add new ones from fresh fan-outs. Annotate changes so trend lines stay interpretable. Prompt trends over time show which questions are rising.
What to do with the results
| Finding | Action |
|---|---|
| A fan-out with no matching page | Brief a new page |
| A persona where you are absent | Create persona-specific content |
| Competitor named in many fan-outs | Study their page with Page Tracker |
| High-volume, high-difficulty prompt | Build a cluster, not a single page |
| Low-volume prompt with high citation rate | Cheap win, write it |
Content gap analysis gives coverage scores to prioritize, and the Action Board turns evidence into a to-do list.
Plan notes
Explore is free with 10 prompts on ChatGPT only, enough to try personas on a few prompts. Essential at $95 a month gives 50 prompts and 6,000 responses. Professional at $245 gives 150 prompts and 18,000 responses. Business at $579 gives 350 prompts and 42,000 responses. Agency plans have unlimited prompts from $199 a month.
In our ranking, other tools handle prompt discovery differently. Scrunch AI's Core plan covers 125 unique prompts, and Gauge runs 600 daily. Volume is not the same thing as relevance.
Who should pick which
| You are... | Recommendation |
|---|---|
| New to prompt research | Three personas, 15 prompts, a manual run |
| A team that wants volume data | Promptwatch Prompt Volumes on a paid plan |
| Selling to several buyer types | Persona stacking |
| An agency | Agency plans, unlimited prompts per client |
| Wanting a quick look | The free GEO Inspector extension |
Start with people, not keywords. Promptwatch lets you hold the personas, the prompts, the fan-outs, and the results in one workspace, which is why it is our choice.