How to measure AI visibility without fooling yourself
Choose a small set of customer questions you can repeat. Record the full answers and compare like with like. That produces a sample you can learn from. It doesn't estimate what every customer sees.
Define the sample before running it
Use separate groups for unbranded buying questions, provider comparisons and questions about your own business. Keep the wording fixed for the comparison. Record date, tool and mode, prompt location, and any known settings. Treat different tools or modes as separate series.
Record three different things
- Mention: Is your business named, and is it being recommended or merely discussed?
- Citation: Does the answer link to your domain or to an independent source about you?
- Accuracy: Does it describe the business and its services correctly?
An illustrative report might say: “Named in two of eight valid answers, cited directly in one, and incorrectly described once.” Those numbers describe the sample, and the three measures can overlap. Report unavailable or failed runs separately; don't count them as absences.
Turn the sample into a decision
Open the citations. Investigate errors you can corroborate. Keep a change log and rerun the same sample after meaningful changes. Compare enquiry evidence separately: ask new customers how they found you, and record their answer even where referral data is missing.
Start with the recommendation check or see how RightView combines AI observations with other evidence.
Download the AI visibility test worksheet.