Build a prompt set around buying decisions
Choose questions your audience could ask while learning, comparing or selecting a solution. Keep branded and unbranded questions separate. A prompt naming your company answers a different question from a request for suppliers in your category.
Preserve the settings
Record the supported source, prompt, market, date and run status. Keep the same settings when comparing runs. If the source or prompt changes, begin a new series or explain the break in the report.
Read the returned evidence
A mention names a brand. A citation identifies a source. An answer may contain one without the other. Inspect the returned text and URLs before drawing a conclusion, and keep failed calls separate from completed checks that did not mention the brand.
Choose a useful response
If answers repeatedly misstate your product, improve the factual information on your own site and relevant public profiles. If a useful question has no answer on your website, consider a well-supported page. Do not manufacture independent recommendations or testimonials to fill a gap.
Report the scope
Say which prompts and sources were checked, then summarize what appeared. A small sample can guide investigation; it does not establish a universal share of AI recommendations.
Read the AI visibility guide for a measurement method. For the monitoring capability itself, see AI visibility tracking.