Your prompt panel is your instrument. If it's lopsided — all branded questions, all one phrasing, reshuffled every month — every chart you build on it will mislead, however clean the dashboard looks.
#Start from real questions
Pull candidates from places where buyers actually speak:
- Sales calls and demos: the questions prospects ask in the first ten minutes.
- Support tickets and community threads: the "how do I…" that precedes a purchase.
- Search Console: question-shaped queries, especially those where you rank just off page one.
- Competitor comparison pages: the criteria rivals choose to emphasise.
Write each as a person would say it, not as a keyword.
#Cover the journey
Group prompts by stage so a change in one doesn't hide in an average:
| Stage | Example |
|---|---|
| Problem | How do I route approvals across teams without email? |
| Shortlist | What are the best workflow platforms for operations teams? |
| Comparison | Brightloom vs Northgrid for multi-team approvals |
| Constraint | Which workflow tools support audit logs and SSO? |
| Purchase | Is Fernhill worth it for a 40-person team? |
Keep branded questions (those naming you) separate from non-branded ones. Branded prompts tell you how you're described; non-branded prompts tell you whether you're found.
#Size it for decisions, not coverage
Twenty to fifty prompts is enough to start. Each additional prompt should have an owner who would act on a change. If nobody will, it's noise.
#Freeze it
A panel is only comparable over time if it stays put. Make changes deliberately:
- Keep the core panel fixed for at least a quarter.
- Add experimental prompts as a separate set so they don't alter the baseline.
- Record the date of any change to either set.
#Repeat, don't trust a single run
Answers vary between runs. Collect each prompt several times per period and report a range, not a point. When a difference is smaller than the run-to-run spread, it's not a finding.
#Record the conditions
Store the engine, country, language, date and any provider settings alongside each answer. Two answers collected under different conditions can't be compared, however similar the prompt.
You can draft a balanced first panel in a minute with our buyer-prompt builder, then replace the generic questions with ones you really hear.