AI search visibility
AI visibility audit: measure where your brand appears and what to improve
Use the audit to choose useful page improvements. A prompt score alone cannot prove traffic or revenue.
The direct answer
An AI visibility audit checks whether a brand appears in answers to relevant buyer questions, how it is described, and which pages support the answer. Use a fixed prompt sample, repeatable settings, and dated evidence. Separate brand mentions, citations, website visits, and qualified enquiries. Each measures a different result.
1. Define the buyer, market, and decision
Begin with a real buying decision. A broad question such as 'best marketing software' mixes audiences and needs. A question about comparing account audit options for a multi-client agency gives the audit a clearer purpose. Use confirmed offers, locations, and buyer types.
Choose the search or answer systems your buyers use, then record the mode being tested. A web-enabled answer, a standard search result, and a model answering without live search are different conditions. Keep them separate in the results.
- Confirm the exact brand name, domain, product names, and likely name collisions.
- Record country, language, date, model or product mode, and whether web search is enabled.
- Use a fresh conversation for each independent test and avoid prompts that steer the answer toward your brand.
- Keep the same core sample for the next review. Add new questions as a separate group.
2. Build a prompt matrix around buyer decisions
Use questions collected from sales calls, support requests, search data, or documented customer research when available. Label internally proposed questions as hypotheses. Your prompt list is a research sample, not a measurement of total search demand.
The following fictional examples show a structure an agency could adapt. They deliberately separate category discovery from a named-brand question. A brand that appears when directly requested has not necessarily earned visibility among buyers who do not know it.
| Decision | Example question | What to record |
|---|---|---|
| Understand the problem | What should an agency check before increasing a client's Google Ads budget? | Advice, named tools, supporting pages |
| Find options | Which tools help agencies review several Google Ads accounts? | Included alternatives and selection criteria |
| Compare approaches | How does an automated account audit compare with a manual agency review? | Tradeoffs, evidence, and missing context |
| Assess a named brand | What data does Example Platform need to review an account? | Factual accuracy and source quality |
| Resolve a constraint | How can I research an agency prospect before account access? | Practical steps and recommended next action |
3. Save the answer and the citation evidence
Record the exact prompt and visible answer, including the cited URLs. Keep evidence private when prompts contain client information. A citation to an unrelated page does not support the same conclusion as a citation to the page that answers the buyer's question.
Score the dimensions separately: correct brand mention, relevant recommendation, factual accuracy, owned-page citation, and third-party citation. Define the rule before reviewing answers. Mark an ambiguous result for review instead of forcing it into a positive score.
- Mention: does the answer name the correct company or product?
- Fit: does it recommend the brand for the buyer and task under review?
- Accuracy: are the offer, audience, access, and limitations correct?
- Citation: does the linked page support the statement beside it?
- Action: can a reader reach a useful next step from that page?
4. Report a sample result with its limits
Repeat the sample under the same conditions because answers can vary. Report the number of prompts, runs, systems, and review dates. Keep a question that produces no AI answer in the record. Do not silently remove inconvenient results.
For an invented sample of 20 questions run twice on one system, 40 answers are observed. If the correct brand appears in 10 answers and an owned page is cited in 4, the sample mention rate is 25% and the owned-page citation rate is 10%. Those rates describe this sample only. They do not estimate market share, impressions, clicks, or revenue.
- Compare the same question set across review periods.
- Show prompt-level changes alongside totals so one repeated question cannot hide a weak topic area.
- Record competitor inclusion only for the same conditions and sample.
- Keep visibility evidence separate from referral traffic and customer outcomes.
5. Fix the page gap the evidence identifies
For Google AI features, normal search eligibility remains the foundation. Google says a supporting page must be indexed and eligible to show a snippet. Also check the site's Search generative AI inclusion setting in Search Console. There is no special AI file or schema requirement. Check crawl access, page text, internal links, and accurate visible content before chasing a new format.
For ChatGPT Search, review OAI-SearchBot access separately from GPTBot training access. ChatGPT-User handles some user-requested visits; it does not control search inclusion. Ask the site owner to verify the intended access in both robots rules and the hosting layer.
Then improve the page that should answer the question. Give the direct answer early. Explain fit and exclusions. Include an original example, decision table, method, or useful template. Cite the primary source for externally checkable claims and show the review date. Update the existing best-fit page when it can serve the need.
| Audit observation | Page work to consider | Validation |
|---|---|---|
| Brand is described incorrectly | Clarify company, offer, audience, and access requirements | Recheck visible facts and repeat the same prompt |
| Useful comparison question has no matching page | Publish a fair comparison with clear selection criteria | Check indexing, citations, and relevant enquiries |
| Cited page gives only a broad claim | Add the method, evidence, example, and limits | Confirm the page supports the cited statement |
| Strong answer page cannot be found | Repair crawl access and relevant internal links | Verify the URL in search tools and repeat the sample |
Sources: Google Search Central: optimizing for generative AI featuresOpenAI: search crawlers and user agents
6. Connect visibility to useful business outcomes
Review the available search reports as well as the prompt sample. Google's 2026 announcement describes dedicated Search Generative AI performance reports in Search Console. Check the current report definitions and the data available for your property before choosing the baseline.
In analytics, inspect identifiable referral visits and what those visitors do next. Some visits may lack a useful referrer, so do not assume the measured group contains every AI-driven visit. Track relevant outcomes such as completed enquiries, qualified leads, or account connections. A citation is evidence of visibility; it is not evidence that anyone clicked or became a customer.
Set a review date for the page change and keep a change log. Compare the same prompt group, search report scope, and conversion definitions. Report what improved, what stayed uncertain, and which buyer question deserves the next piece of work.
Sources: Google Search Central: Search Generative AI performance reports
Put it to work
Keep the evidence beside the decision.
An editable prompt and citation ledger with fields for conditions, evidence, accuracy, and the next page action.
Download the AI visibility audit worksheetCommon questions
What is the difference between AEO, GEO, and SEO?
SEO improves how useful pages can be found and understood in search. AEO and GEO usually describe work aimed at answers and generative search experiences. Define the actual surface, buyer question, and outcome being measured because the labels overlap.
Can an AI visibility audit guarantee more citations?
No. An audit identifies observed gaps and pages to improve. Search and answer systems control which sources they use, and results can vary between runs. Measure changes without promising inclusion.
Does Ad Prophet's Free Growth Read include this AI visibility audit?
No. The Free Growth Read uses public market and buyer evidence for a limited growth read. This guide provides a separate worksheet for running your own AI visibility review. Confirm the scope of any additional analysis with the team.
Sources and method
This guide combines the linked sources with Ad Prophet's editorial methods and original decision aids. Numerical examples are illustrative. They do not describe client results or industry benchmarks.