Audit AI search visibility by recording specific questions, checking the pages that should answer them, and collecting evidence of access, accuracy, and usefulness. An absent mention is an observation to investigate, not proof of a technical failure. No provider can guarantee that a business will be selected or mentioned in an AI-generated answer.
Our AI search visibility overview explains the foundations. This audit turns them into a working record: which question was tested, which page should answer it, what evidence exists, who owns the fix, and when it will be checked again.
1. Define four groups of customer questions
Start with a small set of meaningful questions your business is qualified to answer. Keep their wording in your worksheet so the next review can use the same starting point.
- Branded: your business name with its actual service or contact question. Does the result describe the right business?
- Service: a need the business actually handles, such as help maintaining an existing website. Which page explains that offer?
- Local: the service and a place you genuinely serve. Does the linked content explain coverage accurately?
- Comparison: a buying decision, such as managed website support versus basic hosting. Does your content help someone choose?
These are example categories, not evidence that particular queries have search volume. Use customer inquiries and available search data to choose the actual wording. Include questions that should lead to a service inquiry, not only questions likely to repeat the business name.
2. Record what the search experience actually shows
Review relevant Google results and AI-assisted experiences available to your customers, such as ChatGPT search. Record the date, platform, query or prompt, search mode, relevant location or language context, whether the business appeared, the factual description, and any linked destination. Save a screenshot or result URL where appropriate.
Keep “not mentioned,” “mentioned without a link,” and “linked to an irrelevant page” as different observations. When no AI answer is shown, record that too. Do not label it a ranking of zero. Use a fresh conversation for a repeat observation and note any context you supply; a prompt that gives the answer is not independent evidence of discovery.
A small manual sample helps find problems and ideas. It does not represent every customer's results, establish a stable AI ranking, or prove that a recent edit caused a mention. Answers and available features can change.
3. Verify access and eligibility at the page level
For each important service or resource URL, review the public page and its search status. Record separate findings for the following checks rather than giving the entire site one pass:
- Public response and content: the intended page loads, important facts appear as readable text, and a login, broken script, or error does not hide the answer.
- Search settings: check the canonical, indexing directives, robots rules, and relevant hosting or firewall restrictions. Keep private areas protected.
- Google status: use URL Inspection to review indexed information and, separately, test the live URL. A successful live test does not prove the page is indexed.
- Google AI inclusion: review the property's Search generative AI control, including inherited settings, if visibility in those features is intended.
- Platform-specific access: have the site administrator review relevant crawler rules and network restrictions using each provider's documentation.
Google's generative AI search guidance connects eligibility with normal search requirements, snippet eligibility, and Search Console inclusion. It does not promise selection. OpenAI's crawler documentation separates OAI-SearchBot for search from GPTBot for training; changing one is not a universal setting for the other. Ask the administrator to verify the intended policy rather than disabling security protections broadly.
4. Reconcile business facts and content
Choose an approved source of truth for your business name, public contact details, services, and actual service area. Compare it with the homepage, About page, service pages, contact page, and relevant business profiles. Record the exact conflicting URLs. Do not invent a local office to make a service-area business look closer to a searcher.
Next, read the page that should answer each query. Can a potential customer identify what you offer, who it fits, any important limitations, and the next step? Does the page provide useful specifics instead of broad claims? Are related services and resources connected with normal descriptive links?
Review any structured data against the information visible on the page. Use supported markup for real content; a special “AI visibility” badge or hidden promotional text is not an audit fix. If a useful page already answers the question, improve it before creating a competing version.
5. Use an evidence worksheet
Copy the field list below into a document or spreadsheet, or print this page. Create one row per finding. Mark results pass, fail, or unknown. “Unknown” is useful when access or reliable evidence is missing; do not silently count it as a failure.
| Field | What to record |
|---|---|
| Question and intent | Exact wording; branded, service, local, or comparison; intended customer. |
| Test context | Date, platform, mode, language or location context, and any supplied information. |
| Expected destination | Canonical page URL that should answer the question. |
| Finding and status | Observed issue; pass, fail, or unknown; keep a missing mention separate from an access failure. |
| Evidence | Exact page or report URL, screenshot reference, inspection date, and relevant result. |
| Customer impact | Who is affected and whether the issue blocks access, creates confusion, or weakens an answer. |
| Action and owner | Specific proposed correction, responsible person, and any access or approval needed. |
| Retest | Target date, verification method, result after the change, and unresolved questions. |
Example finding, not a measured client result
Question: “Does this business manage existing websites?” Finding: the service page implies only new builds, while the management page explains existing-site reviews. Status: fail for content clarity, unknown for AI selection. Action: clarify the service description and link to the management page. Retest: verify the public content and link after publishing, then repeat the recorded search observation at the next review.
6. Prioritize fixes by customer impact
- Fix blockers first: important public pages returning errors, unintended indexing exclusions, or broken inquiry paths. Verify the repair on the live site.
- Resolve factual conflicts: wrong services, contact details, or locations. Correct the actual source and relevant profiles.
- Improve incomplete answers: clarify a service, add relevant evidence, explain limitations, and connect the next step.
- Revisit discovery observations: repeat the same sample after allowing time for platforms to process changes. Keep uncertainty in the report.
For each action, define success you can verify: a working URL, consistent contact information, an accurate service answer, or an accepted inquiry. Avoid a combined “AI score” that hides the difference between a technical test and a variable search observation.
7. Measure visibility and business outcomes separately
Review traditional search performance and the reports actually available in your accounts. Google's Generative AI performance report provides impression information for supported Search features and may be absent when there are insufficient impressions. Record the report, date range, and dimensions used. Do not treat an impression as a visit or purchase.
Where available, review relevant referral sessions and confirmed inquiry events in your existing analytics. Keep test submissions, spam, checkout clicks, and genuine customer outcomes separate. Missing referral information is not proof that no AI-assisted discovery occurred.
A useful ongoing review reports what changed, which evidence was collected, what remains unknown, and which customer problem should be addressed next. It cannot guarantee future mentions or attribute every lead to an AI response.
Official platform documentation reviewed October 30, 2026. Search products and reporting can change. The query groups, worksheet, and prioritization method above are Colby Technologies recommendations.
Turn the findings into an ongoing SEO plan
Share your website and the customers you want to reach. Colby Technologies can review the important pages, identify supported fixes, and scope recurring SEO work around relevant content and qualified inquiries.