Key takeaways
- Compare evidence gathered under similar conditions over time, not one favourable answer.
- Keep branded questions separate from unbranded discovery questions.
- Record mentions and citations as distinct signals.
- Save the exact question, date, platform, mode and conditions as a baseline.
- A repeatable, honest record matters more than a single impressive result.
Business owners often want a simple yes-or-no answer about AI visibility. The honest answer is more useful: visibility can shift with the question, platform and exact moment of the check. Measuring it properly separates genuine improvement from coincidence.
What counts as evidence?
Useful evidence includes whether a business is mentioned, whether its pages are cited, whether the details are accurate and whether identifiable visits lead to meaningful actions such as enquiries. Record these separately: a mention is not a website visit, and a visit is not an enquiry.
Build a question set worth tracking
Separate questions that name a company from those that do not. A named question checks how an AI describes a business it already knows about; an unbranded question tests whether the business gets discovered without being named first. Keep the core wording stable between reviews.
Record a baseline
For each check, save the exact question, date, platform, model or mode, and relevant conditions such as whether web search was enabled. Use a fresh conversation for standalone discovery questions where practical. Save the answer text, visible sources and a screenshot so someone else can verify it later.
Useful spreadsheet columns
- Question, purpose and branded/unbranded status.
- Date, platform, mode and location or account conditions.
- Business mention: named, not named or unclear.
- Context: recommendation, example, comparison or another role.
- Citation URL and whether it is your own page or a third party.
- Accuracy: correct, partly correct, incorrect or not assessable.
Track mentions and citations separately
A mention names the business. A citation identifies a source page. A business may be cited without being recommended, or named while the answer links to a directory instead of its own website. Record the context, then check whether a cited page actually supports the claim beside it.
Review consistently
Choose a schedule you can maintain, such as monthly. Repeat the same questions with the same method and do not refresh until you find a flattering answer. Preserve variation and look for patterns across several reviews. If conditions change substantially, create a fresh baseline instead of stretching an old comparison.
Where analytics fit
Search Console and analytics can provide additional context, but they should not be treated as isolated proof that an AI feature caused a change. Track completed work separately too: publishing a page, getting it indexed, appearing in an AI answer and receiving an enquiry are different outcomes requiring different evidence.
Measure progress honestly
End each review by recording what changed, what remains uncertain and the next action. Keep mentions and citations separate, branded and unbranded checks apart, and connect visibility evidence to visits and enquiries without overstating the link.