Measure the answer you actually observed
Start with a small set of questions your buyers might ask. Save the exact wording, product used, date, visible mode and answer evidence. A brand mention and a link to your website are separate observations. Neither alone establishes a visit, inquiry or sale.
The goal is a repeatable record you can inspect, not a universal AI ranking. Results can vary across prompts, products, sessions and time. A useful report preserves that context rather than compressing everything into an unexplained score.
Build a question registry
Give each question a stable ID and version. Separate branded questions from questions that do not name your business. Include the buying task, intended market and language. Choose the platforms and repeat schedule before starting, and record settings you can inspect. Mark hidden or unavailable settings as unknown.
Keep the question set manageable enough to read the actual answers. Repeating a few relevant questions carefully is more useful than collecting hundreds of rows that nobody checks. If you revise a question, keep its previous version so comparisons remain understandable.
Keep collection failures visible
Use separate statuses for a completed answer, a failed collection, a check you did not run, and a search result that presented no AI answer. Declare which observations enter each calculation. A failed request is not evidence that the brand was omitted.
Fictional example: eight checks are planned; six produce completed answers, one fails and one is not run. Two of the six answers mention the brand, and one links to its website. Coverage is 75%, the sample mention rate is 33.3%, and the owned-citation rate is 16.7%. Show the counts beside those percentages. They describe this sample, not market share.
For the workbook’s answer-coverage measure, a completed search check that presents no AI answer remains in the planned-check count but not the completed-answer count. It is excluded from mention and citation rate denominators. Report those no-answer checks separately; they are observed search outcomes, not collection failures.
Read the sources, not just the mention count
Open each cited URL that matters to a finding. Check whether it supports the claim and whether it is your website, a third-party source, or an unrelated reference. Prioritize incorrect services, prices, contact details and other facts that could mislead a buyer.
Record the observation, the source page that needs inspection, a responsible person and a follow-up date. A later change in an answer does not prove that your edit caused it. Google’s guidance on AI search features is useful for understanding its own search environment; other products have different behavior and documentation.
Compare compatible samples
Compare like questions under a consistent collection method. If the question set, platform mix or known context changes, mark the break or report a clearly labeled matched subset. A compatible sample still needs enough observations before you draw a strong trend conclusion.
Keep referral visits and qualified inquiries in a separate report with their own analytics and CRM sources. Use the measurement framework to connect business outcomes responsibly, and the AI search visibility service when you need help interpreting what to change.
If you hire someone, ask for a usable record
A useful audit should let you inspect the exact questions, collection dates, visible platform settings, answer evidence and sources behind each finding. Ask for the completed-answer counts and failed checks as well as the summary percentages.
The handover should separate observed factual errors from hypotheses, identify the page or business fact to investigate, and explain how the next check will be compared. A score without that supporting record is difficult to act on. This workbook gives you a place to retain the evidence.
Take the worksheet with you
Use the blank worksheet for your own review. Any completed examples are clearly labeled as fictional.