Before you start
Choose one brand, one target market, and a concrete research objective. Collect customer questions from interviews, sales notes, support requests, or approved first-party research. Remove personal information before reusing customer wording.
- Output: a versioned list of prompts with intent and context.
- Decide who maintains the portfolio and who reviews the evidence.
- Download the prompt planning worksheet and use one record per question.
1. Group the customer decisions
Organize the questions around decisions your customers make. Discovery asks what options exist. Comparison asks how alternatives differ. Decision questions introduce a constraint such as integration, budget, or location. Support questions concern a product already in use.
Keep these groups separate in reporting. A brand-named support question and an unbranded discovery question have different purposes; averaging them together can obscure the result.
2. Write a question without steering the answer
Preserve relevant context, but remove instructions that force the platform to mention your brand. Put branded questions in a separately labeled group. Do not treat paraphrases of one question as independent evidence of broad demand.
Use the customer’s language where possible. Record a question’s source and the reasoning behind its inclusion rather than claiming it represents popular demand without supporting data.
3. Record the observation settings
Save the exact prompt alongside the platform, market, language, and observation date. Record conversation history or other context when it is part of the research design. Keep settings consistent when comparing brands or periods.
When a setting cannot be controlled, say so. A stated location in a prompt is not evidence that a platform applied that location throughout its response.
Worked example: a fictional scheduling product
Suppose a team wants to understand how small clinics discover scheduling software. “Which scheduling tools support a small clinic with multiple practitioners?” belongs in discovery. “How do Product A and Product B handle recurring appointments?” belongs in comparison.
A follow-up such as “Which option supports reminders in two languages?” introduces a specific constraint. Store it as a follow-up with its preceding context, not as an unrelated standalone question. These examples are a planning exercise, not measured search demand.
4. Review coverage and freeze a baseline
Review the list for missing decisions, repeated wording, and assumptions about the audience. Start with a manageable set that someone can inspect; there is no universal number that makes a portfolio representative.
Give the baseline a version. When questions are added, removed, or materially rewritten, record the change. Report like-for-like comparisons on the common set and show the new questions separately.
Common mistakes
Do not select questions only because they already mention your brand, replace a low-performing prompt without recording the change, or call your sample “all customer questions.” Each of those choices changes what the result means.
- Keep a source and purpose for each prompt.
- Separate failed observations from valid answers.
- Preserve an unchanged comparison set when expanding coverage.
Your next task
Run the initial observations and save the full answers. Use the measurement guide to define numerators, denominators, and exclusions before you calculate visibility.
Common questions
How many prompts should we track?
Choose a set you can inspect and maintain that covers your defined decisions. The count alone does not establish representativeness; explain what your portfolio includes and omits.
Can we change the list later?
Yes. Version the change and preserve a common comparison set so a portfolio change is not mistaken for a performance change.
Sources and methodology
Digraph publishes this guidance and provides AI visibility software. Worked examples are illustrative unless explicitly identified as measured results. A research sample does not establish total market coverage or guarantee future recommendations.
Continue the handbook
Capture a baseline and make comparisons with consistent denominators.
Run your first visibility audit