An AI visibility baseline is a repeatable sample of questions and observed answers, not a universal ranking. Record the system, question, date and cited sources so later comparisons have context.
Define the discovery situations
Create questions around real user needs, alternatives and constraints. Include category-level requests and specific use cases. Avoid selecting only questions that already mention your brand; those test recognition rather than discovery.
Keep the sample focused enough to repeat. For an illustrative journaling app, questions about private reflection, short daily entries and cross-device use may represent different decision contexts.
Record more than a mention
Note whether the app appears, whether the description is accurate, which alternatives appear and what sources are cited. An inaccurate recommendation is not a clean success. A useful citation without a direct recommendation can still reveal discoverable evidence.
Distinguish ordinary search referrals from sampled answer observations. They measure different things and should not be combined into a single unexplained score.
Control what you reasonably can
Use a consistent question set and document relevant settings or session context. Repeat observations when practical because outputs can vary. Do not claim that a small sample represents every user, geography or version of a system.
The baseline should surface content gaps and factual inconsistencies, not create an illusion of precision.
Turn findings into useful work
If your app is described incorrectly, improve clear owned product information. If relevant use cases lack supporting pages, create substantive explanations. If sources are outdated, identify which information you control and what requires an external correction.
Keep a change log and review results alongside qualified traffic and inquiries. The value of monitoring is the decisions it informs. A report that only counts mentions without investigating accuracy or user relevance leaves most of that value unused.