Prompt set
Updated 2026-09-15
A prompt set is the fixed list of questions a brand tracks across AI engines. It is the measurement instrument: every rate you report is a property of this list as much as of your content.
How to build one
A workable set covers five intent classes — category discovery, head-to-head comparison, use-case or integration questions, pricing, and switching — with several phrasings each, in the words buyers use rather than the words marketing uses. Thirty to forty prompts is the usual range: wide enough to see a pattern, small enough to sample repeatedly. Branded prompts belong in the set but should stay a minority, because they measure recognition rather than discovery.
Illustrative split (example set)
A 35-prompt set: 10 discovery, 8 comparison, 7 integration, 5 pricing, 3 switching, 2 branded. The two branded prompts show mention rate near 1.0 and tell you almost nothing; the ten discovery prompts carry the coverage number that actually moves.
Common mistakes
- Editing the set between cycles and then reading the change as progress — the instrument changed, so the trend is not a trend.
- Filling the set with branded prompts, which inflates every rate.
- Writing prompts in vendor language, which measures a question nobody asks.
Frequently asked questions
- When is it acceptable to change the prompt set?
- When the market changes or a whole intent class is missing — and when you note the change in the report and treat the next cycle as a new baseline rather than a continuation.
- Should competitors appear in prompts?
- In a few, yes: "alternatives to X" and "X vs Y" prompts capture switching demand and are usually easier to win than open discovery prompts.
