THE USEFUL TAKEAWAY
Use a documented question set and repeated observations. Keep visibility, accuracy, referrals, and qualified inquiries as separate measures.
Define the question before the score
“Are we visible in AI?” is too broad to measure on its own. Choose a customer, a need, and a discovery context. A homeowner comparing emergency plumbing options asks a different question from an agency researching software.
Write a manageable set of realistic questions. Include service and location where appropriate, but do not fill every question with your own brand name. Branded prompts can test factual accuracy; unbranded prompts can explore discovery. Keep those groups separate so a report does not confuse knowledge of your name with a recommendation to a new customer.
Record the testing conditions
For each observation, record the exact prompt, date, engine or product, available model label, account state, relevant location or language settings, and whether search was enabled. Save the answer and source URLs where possible. If a setting is unknown, record it as unknown.
Repeat questions rather than assuming one response is stable. You do not need to generate an enormous report to learn something useful. A small, consistently documented set is often easier to interpret than hundreds of untracked prompts. Follow provider terms and avoid an automation method that your tools are not authorized to use.
Choose clear fields
Label whether the correct business was mentioned, whether a source was cited, whether the context suggested the business as an option, and whether the facts were accurate. Record uncertainty instead of forcing every response into a binary answer.
If you calculate a mention rate, state the numerator and denominator. “Four appearances across twenty recorded responses to this question set” is more informative than “20% AI market share.” Your sample does not represent every user, question, or location. Do not compare two reports unless their question sets and methods are sufficiently similar.
Connect observations to business outcomes
Track relevant website referrals where you can identify them, then examine qualified inquiries and customer feedback. Referral data can be incomplete: some visits arrive without a useful referrer, and some people search for your name later. Ask new customers how they first heard about you when that question fits your workflow.
Search Console offers platform-specific reporting for your own website. Review the current official reporting documentation and the labels available in your account rather than assuming every AI surface is measured in the same report. No single analytics source sees the whole customer journey.
Interpret changes carefully
Keep a change log for important content, profile, and technical updates. Compare later observations with the baseline, but remember that product behavior, competing pages, and user context can also change.
- Report the method and observation dates.
- Distinguish visibility from correctness and conversion.
- Keep examples of inaccurate or ambiguous responses.
- Describe correlations without claiming an experiment proved causation.
The useful result is a decision: correct an outdated fact, improve a service explanation, investigate a cited source, or keep observing. A score that cannot lead to a sensible next action is less valuable than a smaller, transparent record.
Sources & further reading
Primary documentation for the platform-specific points in this guide. Practical workflows and hypothetical examples are Rankiful’s editorial guidance.
Platform features and policies change. Check the current official documentation before making technical changes.