Track buying decisions, not just variations of your own brand name.
Collect the questions before writing prompts
Start with the language customers already use. Review sales notes, onboarding questions, support requests, site search, and interviews. Record the problem, the constraints, and the alternatives mentioned. A useful prompt is grounded in a real decision; a polished sentence invented in a marketing meeting may represent nobody.
Separate explicit brand questions from category discovery. ‘Does Acme connect to my accounting system?’ tests a different stage from ‘Which accounting tools work for a small construction business?’ Both matter, but a brand-heavy set can make visibility appear healthier than discovery really is.
Build a coverage matrix
Use rows for decision stages: learning, shortlisting, comparing, validating, and purchasing. Use columns for your important audiences or use cases. Add questions where you have customer evidence, and mark empty cells as research gaps rather than filling them with guesses.
For a hypothetical project-management product, one row might compare agency tools, another ask about external client access, and a third examine migration from spreadsheets. Preserve the relevant constraints. If price, geography, team size, or compliance matters to the buyer, include it explicitly.
- Store a prompt ID, exact wording, topic, intent, audience, language, market, and evidence source.
- Mark brand-specific prompts separately from unbranded discovery prompts.
- Avoid giving one topic extra weight merely because it has more paraphrases.
Keep a baseline and an exploration set
A fixed baseline makes changes easier to interpret. An exploration set lets you test new products, emerging objections, and unexpected query patterns. Report them separately: adding easier questions to a baseline can improve the headline number without changing the brand’s position in any original question.
Version your prompt set. When a question changes, create a new version and retain the old wording. Record why you changed it. This turns an unexplained chart discontinuity into an understandable research decision.
Review with someone close to the customer
Ask a salesperson, account manager, or support specialist to challenge the set. Which questions sound artificial? Which decision constraints are missing? Which prompts assume customers know your category terminology? Replace the weakest questions before expanding coverage.
A prompt panel is a sample you designed. Unless you have separate, documented evidence of real usage, it does not measure how often people ask those questions. Keep observed visibility and estimated demand in different columns.
