Define the measurement decision
Treat the prompt set as a research instrument. Changes require version notes and should not silently alter the denominator of a trend.
Measurement method
| Step | Action |
|---|---|
| 1 | Create Prompt Groups for category, use case, comparison, product, and risk questions |
| 2 | Set a minimum evidence reason for every prompt and remove orphan questions |
| 3 | Balance groups so one topic does not dominate the overall view |
| 4 | Record additions, retirements, and wording changes as panel versions |
| 5 | Review failures, duplicates, and low-action prompts before expanding the set |
Worked calculation or observation
A 200-prompt panel with 120 near-identical tool comparisons can make one topic look like the whole market. A smaller balanced panel gives a more interpretable operating signal.
Evidence to retain for managing multiple AI search prompts
Keep these fields with the decision:
- metric name
- numerator
- denominator
- scope
- platform
- Prompt Group
- success state
- observation date
- formula version
Limits and UnderAI's role in managing multiple AI search prompts
Do not compare trend lines across materially different prompt panels without labeling the break. Prompt text changes the measurement.
UnderAI supports Prompt Groups and immutable prompt text after creation, which encourages versioned additions rather than silent edits.