Every AEO metric, from mention rate to share of voice to citation share, depends on the prompt set you measure. A set that's too small, too branded or worded unnaturally gives you confident-looking numbers that don't reflect what buyers actually see. This guide covers how to design a prompt set that's representative, balanced and stable enough to track over time.
Principles of a good prompt set
- Representative: it reflects the real questions buyers ask, at every stage.
- Balanced: weighted toward the prompts that matter most, not the ones you already win.
- Natural: worded the way people type into AI assistants.
- Segmented: tagged so you can analyze by stage, persona and product.
- Stable: a fixed core that lets you compare over time.
Step 1: Source real questions
Start from real buyer language: sales calls, support tickets, onboarding surveys, search queries, communities and the assistants' own follow-up suggestions. See how to find the questions your buyers ask AI.
Step 2: Cover the journey
Aim for coverage across stages. A typical distribution for a B2B or considered-purchase brand:
| Stage | Example | Share of set |
|---|---|---|
| Problem / education | "How do agencies track billable hours accurately?" | 10–15% |
| Category shortlist | "Best time tracking software for agencies" | 20–30% |
| Use case / constraint | "Time tracking app that integrates with QuickBooks for a 10-person team" | 20–30% |
| Comparison | "[You] vs [Competitor]"; "[Competitor] alternatives" | 15–20% |
| Validation | "Is [you] worth it?"; "[you] reviews" | 5–10% |
| Pricing / decision | "How much does [you] cost for 15 users?" | 5–10% |
Adjust for your business. Ecommerce will have more product attribute prompts; local businesses more "near me" style prompts.
Step 3: Balance branded and unbranded
- Unbranded prompts (no brand names) show whether you make the shortlist. These should be the majority, typically 60 to 80%.
- Competitor-branded prompts ("alternatives to [competitor]," "[competitor] vs") show whether you appear in rivals' conversations.
- Your branded prompts ("[you] pricing," "is [you] legit") show how you're described and whether facts are accurate.
A set that's mostly branded prompts will flatter your mention rate and tell you little about the market.
Step 4: Write prompts naturally
- Full sentences, as typed into a chat
- Real constraints: team size, industry, budget, location, integrations
- Plain language, including casual phrasing where buyers use it
- No leading wording ("Why is [you] the best…")
Compare:
- Keyword-style: "agency time tracking software"
- Natural: "What's the best time tracking tool for a small design agency that bills clients hourly?"
Step 5: Add variants thoughtfully
AI answers can be sensitive to wording. For your most important intents, include two or three phrasings, such as a formal one, a casual one and one with a different constraint, and analyze them as a cluster. This reduces the risk that one quirky phrasing distorts results.
Step 6: Tag everything
Give each prompt tags for:
- Stage
- Persona (end user, buyer, technical evaluator, executive)
- Product line or use case
- Branded / competitor-branded / unbranded
- Region or language
- Priority (business value)
Tags let you answer questions like "How's our share of voice on unbranded shortlist prompts for the enterprise persona on ChatGPT?"
Step 7: Decide on size and runs
- Size: 50 to 150 prompts is practical for most brands. Large brands with many product lines may track several hundred, segmented.
- Runs: three to five runs per prompt per platform per cycle, because answers vary. See why AI answers change every time.
- Platforms: ChatGPT, Google AI Overviews and AI Mode, Perplexity and Gemini at minimum; add Claude, Copilot, Grok and Meta AI as relevant. See which AI platform to prioritize.
Step 8: Maintain it
- Keep a stable core of prompts unchanged for trend reporting.
- Review quarterly for new products, competitors, features and buyer language.
- Add prompts in a separate "new" segment for a cycle before merging into the core.
- Log every change with the date and reason, so shifts in metrics can be explained.
Example: a starter set for a clinic scheduling tool
- "What's the best scheduling software for a small physical therapy clinic?"
- "Appointment booking software for clinics that sends text reminders"
- "HIPAA-compliant online scheduling for a dental practice"
- "Clinic scheduling software that integrates with [EHR name]"
- "[Competitor] alternatives for a multi-location clinic"
- "[You] vs [Competitor] for a 5-provider practice"
- "How much does clinic scheduling software cost per provider?"
- "How do clinics reduce no-shows?"
- "Is [you] easy to set up?"
- "[You] reviews from physical therapists"
The bottom line
Your prompt set determines what your AEO metrics mean. Build it from real buyer language, cover every stage, weight it toward unbranded prompts, write naturally, tag thoroughly, run each prompt several times and keep a stable core. AEO-Rank helps you build and run prompt sets like this across AI assistants, so the numbers you report reflect what your buyers actually see.
On estimating how often each prompt is asked, see AI prompt volume.
