How to run an AI visibility audit your team can use

1. Give your AI visibility audit a specific purpose

An AI visibility audit starts with a business decision. Perhaps buyers cannot find your product for a new use case. Perhaps assistants describe an outdated service. Write down the question your team needs to answer before collecting screenshots or calculating a score.

Define the audience, country, language, and product scope. Then choose which assistant experiences to observe. A consumer chat interface and a search feature can produce different evidence. Record the exact experience you use so another person can repeat the work.

Agree on a small initial scope that someone can maintain. One market and one product category are often enough for a first cycle. Our guide to GEO, AEO, and SEO explains how this work connects with your wider search program.

2. Build the question set from buyer conversations

Collect questions from sales calls, support tickets, on-site search, and interviews. Remove personal information before using any material. Keep the original wording where it helps preserve the buyer’s intent, then group questions by the decision they support.

For a payment provider, useful questions might concern supported hardware, installation effort, or international settlement. “Which provider is best?” gives less context than “Which payment systems support our existing vending machines in Spain?” The second question helps you inspect a specific buying requirement.

Keep branded accuracy questions separate from category discovery questions. Otherwise, a large number of easy branded mentions can obscure weak category coverage. Read our guide to high-intent questions for a practical way to prioritize this list.

• Discovery: which options address the buyer’s problem?
• Evaluation: which capabilities, costs, and limitations matter?
• Selection: which provider suits the stated requirements?
• Accuracy: are product names, locations, and services correct?

3. Capture enough context to interpret each answer

Save the question, date, assistant, visible model label, locale, and relevant session conditions. Preserve the complete answer and its source URLs. A screenshot can support the record, but searchable text makes later review easier.

Use separate fields for a brand mention, a recommendation, and a link to your website. Also record the answer’s qualifications. “Consider this provider if you need local support” means something different from an unconditional recommendation.

Repeat selected questions under the same documented conditions. Keep each observation, including unexpected answers. Repetition can reveal variability, but your sample still describes only the runs you observed. It cannot establish how every buyer sees your brand.

4. Check the source pages and technical access

Open the sources cited in each answer. Check whether they actually support the description, and note publication or review dates. Compare those sources with your product documentation and service pages. Look for missing conditions, inconsistent names, and outdated facts.

For Google’s AI search features, Google’s eligibility guidance keeps established search requirements in place. A supporting page needs to be indexed and eligible for a search snippet. Special AI markup does not guarantee inclusion.

When a page appears inaccessible, use Search Console’s URL Inspection tool to investigate your own URLs. Compare indexed information with the live page. A successful live check confirms access conditions; it does not promise indexing.

5. Turn each finding into a small, owned task

A useful finding describes the observed problem and its business consequence. “Publish the supported-device list on the integration page” gives a team something concrete to do. Add the answer record that exposed the gap and the person who can verify the facts.

Prioritize incorrect facts that could affect a buying decision. Next, address missing explanations on important pages. Finally, consider wider content opportunities. Avoid treating every absent mention as a content problem: some questions may fall outside your offer.

Use our citable-page checklist when turning a finding into a publishing brief. It helps keep definitions, evidence, and limitations close together.

• Evidence: the saved answer and source behind the finding.
• Action: the page or process that needs a change.
• Owner: the person responsible for approval and delivery.
• Completion: the facts, links, or access conditions to confirm.
• Review date: when the team will sample the question again.

6. Work through a concrete audit finding

Imagine an assistant says a fictional provider supports only new machines. The product team confirms that an approved adapter also supports certain older models. The current website describes the adapter in a PDF but omits it from the integration page.

The task is to publish the supported models, installation conditions, and documentation link on that page. A product owner approves the list. An editor explains the limits. A developer ensures the page is reachable from the main product navigation.

At the next review, check the corrected page first. Then repeat the relevant questions and record any response changes. If an assistant still repeats the old statement, keep that result. Publishing the fix and observing its adoption are separate milestones.

7. Finish with decisions and a repeatable review

Keep a stable core of questions between cycles. Add new questions as a separate group, and record any change in assistant settings. This makes it easier to distinguish a content change from a change in how you collected evidence.

Summarize completed actions, unresolved issues, and the next decision. Use clear visibility measures to explain progress without overstating causation. The audit earns its value when it helps the team correct facts and improve useful pages.

Start with one category, assign an owner, and schedule the next review. If you need help defining that scope, a private AI visibility assessment can turn your initial questions into a focused plan.