What Your AI Visibility Report Should Be Able to Prove

By Melinda Starbird — October 12, 2026

A report says AI systems visited your website thousands of times. Another gives your business a visibility score. Both numbers might be useful. Neither tells you, on its own, whether a customer received an accurate recommendation or booked anything.

If you are spending money to improve how your business appears in AI, you should be able to ask what was observed and get a clear answer.

Start with the event behind the number. Did software request a page? Did an answer mention your business? Was your website cited? Did a person arrive, send an inquiry, or complete a purchase?

A request is one piece of evidence

Automated traffic has different purposes. Cloudflare’s documentation distinguishes behaviors such as search crawling, model training, user-directed agent visits, and transactions. It also describes methods used to validate bot identity. [1]

That distinction should carry into reporting. An identified crawler requesting a page is evidence of access. Establishing that the content appeared in a customer-facing answer requires evidence of the answer itself. Establishing a sale requires a transaction record.

Before accepting a large request count, ask which requests were included, how the sender was identified, and whether errors, repeat attempts, internal tests, or monitoring were excluded. Keep the dates and timezone attached to the figure. Without those definitions, a comparison with last month may tell you very little.

Human traffic needs definitions too. Visitors, sessions, and page views describe different things. More page views may reflect more interest, repeat visits, or a person struggling to find an answer. The count alone does not establish the cause.

Test the question a customer would ask

Searching for an exact business name tests whether a system can locate or identify that business. Asking for a service in a town tests whether it appears among relevant choices. Those are different tests and should be reported separately.

A useful report retains the query, platform, date, location assumptions, and the answer or source evidence. If a business is mentioned, check whether the description is correct. The wrong location or an obsolete service can matter more than the mention itself.

Failures also need their own place in the report. A timeout is an incomplete test. It does not establish that the business was absent from the results. A source that cannot be accessed should remain unavailable rather than turning into a confident score.

Those distinctions are my recommended minimum for evaluating a report. They let an owner see what needs fixing and stop missing evidence from becoming a story about performance.

Keep accuracy and outcomes visible

Google says appearances in its AI search features are included within Search Console’s overall Web search reporting. That makes it important to understand the scope of a dashboard before treating its totals as a separate AI channel. [2]

For a small business, I would look at several measures together: unresolved identity conflicts, completeness of important service information, observed answers to relevant questions, and completed customer actions. The measures should remain separate so that improvement in one does not disguise a problem in another.

An increase in citations is worth investigating. So is a correction that stops customers from arriving at the wrong address. An inquiry reaching the owner is a different milestone from a confirmed appointment. A report should make each visible on its own terms.

MiddleVerse should be held to that standard too. Verified business identity needs evidence, and claims about agent activity need a clear account of what actually occurred.

When someone presents an impressive score, ask for one example you can inspect. Read the customer’s question, the answer, and the supporting record. Then follow the next step. That will tell you what the score leaves out.

Sources [1] Cloudflare, Verified bots https://developers.cloudflare.com/bots/concepts/bot/verified-bots/

[2] Google Search Central, AI features and your website https://developers.google.com/search/docs/appearance/ai-features