An AI visibility audit should give you a defined set of buyer prompts, a log of the answers and sources observed, a technical eligibility check for important URLs, and a prioritised list of fixes with owners and retests. Ask for enough evidence to trace each recommendation back to a particular answer, URL or technical check.
This checklist helps an agency brief the audit and review what comes back. Use it to agree what the team will measure, how it will record evidence and which decisions the report should help you make.
What should an AI visibility audit include?
An AI visibility audit should include a scope sheet, an answer log, a URL-level technical review and an action backlog. Ask the auditor to hand over the evidence alongside the summary so another person can check the findings.
Scope sheet. Name the brand and its aliases, the products or services in scope, key URLs, buyer groups, markets, search surfaces, competitors and dates.
Answer log. Save the exact prompts, run settings, observed answers, brand mentions, source URLs and links to the captured evidence.
Technical review. Record access, crawl and indexability findings by URL and surface. Include the specific status, directive or inspection that supports each conclusion.
Action backlog. Give each finding an owner, priority, affected URL and acceptance check. Include any dependency, such as a client-approved product fact or a developer with access to the CDN.
Ask the auditor to label three layers throughout the report. Eligibility describes what a particular platform can fetch or consider. Observations describe what appeared in a recorded answer. Hypotheses suggest what might explain a pattern and say how the team could investigate it. That separation helps you review the scope of an AI search visibility engagement.
How should an auditor select prompts, markets and answer engines?
Start with the decisions buyers make in the markets you serve. Select prompts for discovery, comparison and verification, then choose search surfaces your audience is likely to use. Agree the scope before collecting answers.
For a UK software buyer, a discovery prompt might ask which tools support a particular workflow. A comparison prompt can name two products a buyer would plausibly shortlist. A verification prompt can ask whether a named product supports a specific integration. Add local terms only where they reflect the buyer’s question. Keep the same prompts when comparing runs and record any wording changes as a new sample.
Record the surface and mode, for example ChatGPT with search enabled or Google AI Mode. Log the requested country and language, run date, session conditions and any location limitations. Record whether the surface returned an answer, showed sources or was unavailable. Keep any repeat-run counts separate by surface and prompt group so a reviewer can see what the sample actually covers. Our guide to tracking brand mentions explains how to carry the same prompt set into later monitoring.
How do you check whether a brand mention or citation is accurate?
Read the answer and open the exact cited URL. Check that the answer refers to the right business and that the cited page supports the specific claim. A relevant source can still fail to support a claim about price, availability, a feature or a service area.
Record mentions and citations in separate fields. A mention is a reference to the brand in the answer. A citation is a source link shown by that surface. Record the linked URL, the claim it appears to support, the passage you found and whether it supports the claim, contradicts it or leaves it unresolved. Mark an uncited claim as such. Note redirects or inaccessible pages and preserve the date and evidence reference.
Compare the claim with an authorised current reference for the brand, such as its product documentation or public service terms. Keep a recurring process for reviewing these discrepancies if the audit will become AI brand monitoring.
Which crawl and indexability checks belong in an AI search audit?
Check whether each priority URL resolves to the expected page, can be fetched by the relevant crawler and is eligible for the surface being assessed. Record the final HTTP status and URL, redirect chain, canonical target, robots rules, meta robots or X-Robots-Tag directives, and the presence of the main content in the returned or rendered page. Flag CDN challenges, authentication requirements or errors with the affected URL and date.
For Google AI Overviews and AI Mode, Google says supporting pages must be indexed and eligible to show a Search snippet. With property access, use URL Inspection to compare Google’s indexed information with the live test, including canonical and crawl details where available. Record when indexed information is unavailable. Eligibility is a technical condition, and actual inclusion is a separate observation. Google AI features guidance and URL Inspection documentation
For ChatGPT search, check the site’s treatment of OAI-SearchBot in robots.txt and at the edge. OpenAI documents OAI-SearchBot for search and GPTBot for potential model training, with independent controls. Keep the site owner’s training preference explicit in the audit. Use authorised server or CDN evidence if you need to distinguish a configured allowance from a successful crawler request. OpenAI crawler documentation
Our guide to appearing in ChatGPT search covers the broader visibility work. In the audit, keep confirmed technical findings separate from possible reasons an eligible page was absent from an answer.
How should an audit turn findings into a prioritised action plan?
Turn each finding into one task with evidence, an affected URL, an owner and a retest. Start with confirmed issues affecting important buyer journeys. Give lower priority to sparse or ambiguous observations until you can repeat the test or find stronger evidence.
Illustrative audit extract
This fictional example shows what a handover row could contain. The answer, URL and evidence reference are placeholders.
| Field | Illustrative entry |
|---|---|
| Prompt | Does Example Agency provide weekend onboarding for UK clients? |
| Run context | UK English, ChatGPT Search, one baseline run. |
| Observed answer and citation | The answer says weekend onboarding is included and cites the onboarding page. |
| Evidence link | Placeholder for the saved answer and the cited page passage, record E-014. |
| Affected URL | https://example-agency.example/onboarding |
| Eligibility and hypothesis | A browser opens the cited URL. The page says onboarding fits your schedule. Whether that wording influenced the answer is a hypothesis. Record crawler eligibility separately. |
| Priority and owner | High if the client confirms weekday-only onboarding. The service owner confirms the policy and the content owner corrects the public page. |
| Acceptance and retest | Verify the corrected public copy and rerun the same prompt. Record any later change in the answer separately. |
The acceptance check proves the page correction. The repeat run measures what the selected surface says after the change. Record the original and new answers side by side with dates, including cases where the answer stays the same. Use the backlog to track completed work and keep open questions visible for the next audit.
