AI visibility guide
What AI Visibility Means, and How to Measure It
Separate technical access, accurate descriptions, mentions, citations, and referrals so an AI visibility report answers a specific business question.
On this page
AI visibility is a broad label for how a business or its content appears in assistant answers and AI search results. To measure it usefully, specify the system, the questions, and the outcome you want to observe.
A technical check, a correct description, a linked citation, and a customer inquiry measure different things. Keeping them separate makes a report easier to interpret and a problem easier to investigate.
Start with the question you need answered
Use a working framework like this:
| Area | What you can check | What it does not establish |
|---|---|---|
| Technical access | Whether the relevant system can retrieve eligible public content | That it will select the page |
| Description accuracy | Whether a sampled answer gets the business facts right | That all users see the same answer |
| Mentions and recommendations | Whether the business appears in a defined set of responses | Its share of every real customer query |
| Citations | Whether a response links to the site | That someone followed the link |
| Referrals and inquiries | Visits or business actions you can observe and attribute | The cause of every unattributed conversion |
Choose the row that matches the business problem. If an answer has the wrong service area, start with the claim and its source. If useful pages cannot be retrieved, start with access. If the concern is sales, mentions alone will not answer it.
Technical eligibility is one part of the work
Check page availability, indexing controls, important text, and the links that lead to it. Use the documentation for the particular system you care about.
For a page to appear as a supporting link in Google’s AI Overviews or AI Mode, Google says it must be indexed and eligible to show a snippet. Eligibility does not guarantee selection. Google’s AI features documentation, checked September 13, 2026.
Do not generalize one platform’s rules to every assistant. Record which consumer a check concerns and what it can establish.
Build a repeatable answer sample
Choose a manageable set of questions drawn from the work you sell or the information you publish. Include both factual brand questions and relevant discovery questions, and report those groups separately.
Keep a record of:
- The exact prompts.
- The product and mode used, including whether search was available.
- The testing date and relevant location or language settings.
- The complete answers and cited URLs.
- Whether each request succeeded.
Repeat under comparable conditions. Changing prompts, products, or settings changes the test, so note it when you do.
For an illustrative calculation, suppose a business appears in six of ten completed responses. Its observed mention rate for that sample is 60 percent. If two additional requests failed, report those failures separately. Do not count them as ordinary responses with no mention, or describe the sample as 60 percent of all customer searches.
Assess the answers before aggregating them
Read each mention. An answer that names the business but assigns it the wrong service is different from an accurate recommendation. A citation to an irrelevant page is different from one that supports the answer.
Record those distinctions before combining results. A single grade can hide the difference between an access problem and an inaccurate claim, even though they need different fixes.
For description errors, the business description guide provides a review worksheet.
Connect observations to site work
Choose changes that address findings you can explain. Clarify an ambiguous service description. Repair a broken public page. Correct an outdated fact. Improve a procedure that omits a necessary step.
Save the original finding, the change, and the follow-up result. A later favorable answer is an observation; it is not proof that your edit caused a wider increase in recommendations.
Track useful visits and inquiries with the measurement already appropriate for your business. Attribution may be incomplete, so keep that limitation visible alongside the numbers.
A useful report ends with specific findings, their evidence, and the next checks. It should let a site owner distinguish a problem they can fix from an outcome they can only sample.