TL;DR. A prompt group is the AI Search equivalent of a keyword cluster, built around intent and scenario, not search volume. AI engines deduce intent with high precision. The shorter and more scenario-specific the prompt, the more stable the output. The starter structure: 3 intents × 2 scopes × 2 visibility-hygiene states = 12 trackable prompts per segment.
The 3×2×2 starter group (12 prompts)
| Dimension | Variants | Why it matters |
|---|---|---|
| Visibility hygiene | Branded vs. non-branded | Branded-only = false sense of security. Non-branded is the real battle. |
| Intent | Information / Consideration / Transaction | Same core question, different buying stage. Each pulls a different answer set. |
| Scope | Broad segment vs. niche-in-segment (or geo) | Niches are where small brands can dominate. Always test both. |
Worked example: Mint Mobile (telecom, US)
Segment = mobile phone plans. Niche = prepaid plans. Same core question, three intents, branded and non-branded sides.
| Intent | Scope | Non-branded prompt | Branded prompt |
|---|---|---|---|
| Information | Segment | mobile phone plans | mobile phone plans Mint Mobile |
| Consideration | Segment | best mobile phone plans | best mobile phone plans Mint Mobile |
| Transaction | Segment | mobile phone plans price | mobile phone plans price Mint Mobile |
| Information | Niche (prepaid) | prepaid mobile phone plans | prepaid mobile phone plans Mint Mobile |
| Consideration | Niche (prepaid) | prepaid best mobile phone plans | prepaid best mobile phone plans Mint Mobile |
| Transaction | Niche (prepaid) | prepaid mobile phone plans price | prepaid mobile phone plans price Mint Mobile |
How to read it: the non-branded column tells you whether the AI considers Mint Mobile at all in the category. The branded column tells you whether the AI knows the brand exists when asked directly. Compare the two. That gap is the entity deficit you fix in Steps 3–5.
Where to source prompts (in order of speed)
| Source | Effort | Notes |
|---|---|---|
| Rankscale Prompt Research | Low | Run on the seed domain, returns intent clusters |
| Rankscale Find Terms | — | Free after Page Audit — surfaces queries the pages already address |
| Google Search Console (regex) | Low | ^(what|how|which|best|top|why|where|compare|vs|should) |
| Sales / demo call notes | Medium | Top 5 questions prospects ask before buying |
| G2 / Capterra / Reddit threads | Medium | Real questions, real language, real objections |
Pick the right AI engines per ICP
| Selling to… | Track these engines |
|---|---|
| Enterprise | ChatGPT, Claude, Copilot |
| SMB | ChatGPT, Gemini, Google AI Mode, Copilot |
| Startups / dev tools | ChatGPT, Claude, Gemini (+ Grok for US) |
| Consumers | ChatGPT, Google AI Overviews & AI Mode |
| Europe (esp. France) | Add Mistral |
Frequency: daily for the first 2 weeks to build a baseline, then potentially decrease to bi-daily or weekly for reporting. Bi-daily during product launches or PR pushes. Monthly for general trend detection with numerous similar prompts.
Do this now:
Pick one segment. Build the 3×2×2 prompt group for it (12 variations). Set 2–3 AI engines based on the buyer type. Run daily for 2 weeks.
Full course: Module 2: Setting up Prompt Groups.
Start improving your AI visibility today with Rankscale.
Get started