At a glance
Audit AI search visibility with a fixed set of real buyer questions, record the platform and session conditions, then score mentions, citations, accuracy and referral visits separately. Repeat the sample over time and fix ordinary indexability and content gaps before chasing special AI tactics.
A screenshot of your brand in an AI answer is satisfying. It is also a weak basis for a marketing decision. The answer may change in another session, and a favourable mention tells you little about whether the right buyer learned anything useful.
Our view: an AI visibility audit earns its place when it improves the information buyers can find about you. A rising mention count is worth investigating, but it should not become the objective of the editorial team.
That changes the purpose of the exercise. Instead of trying to win every answer, you are looking for a pattern: where your business is described accurately, where useful evidence is missing and which gaps deserve attention. The checks below give that judgement a repeatable basis.
Choose questions from real buying and learning needs
The question set is where an audit can quietly become a vanity exercise. If you choose questions because your brand already appears in the answers, the result will look reassuring. Questions from sales conversations, customer interviews and Search Console offer a more useful starting point: what do people need to understand before considering you?
Do not fill the set with prompts containing your brand name. Those can test factual accuracy about the company, but they do not show whether someone unfamiliar with you would discover it.
Keep a separate group for branded checks: what the assistant says you do, whom you serve and how it describes your products. Accuracy and discovery deserve different scorecards.
Record the conditions
A change in the answer is only useful evidence if you know what else changed. Was it the wording of the question, the platform or the conversation leading up to it? Recording the exact prompt, date, visible model or mode, language and location context helps you avoid attributing every difference to your latest content edit. Saving the answer and its sources makes the review possible later.
Repeat a manageable sample. A single run can be affected by variation in retrieval and response generation. A stable testing routine will not eliminate that variability, but it makes changes easier to interpret.
Use the same prompt set for a reporting period. Add new questions deliberately and mark them as new rather than blending them into a historical percentage.
Separate the things you observe
- Mention: the brand appears in the answer.
- Citation: the answer links to a page you control.
- Accuracy: the description matches current facts.
- Recommendation context: the answer explains when the brand is suitable.
- Referral: a measurable visit arrives from the platform.
These distinctions matter because a flattering answer can still be a poor result. Imagine the assistant recommends your company for a service you no longer offer. The mention count has gone up, but a buyer has been misled. We would put correcting that description ahead of collecting another positive screenshot.
Check the website before commissioning an AI fix
Google’s guidance for AI features says pages must be indexed and eligible to appear with a snippet to qualify as supporting links. It also says traffic from those features is included in Search Console’s overall Web reporting. That report does not, by itself, measure visibility across every AI answer service.
Check indexability, useful internal links, clear text and accurate public information. Make important product and organisation facts easy to find. Our GEO guide explains the broader relationship between useful content and AI-mediated discovery.
Do not assume a new file or a special schema label will create citations. Repair ordinary publishing problems first: inaccessible pages, contradictory descriptions, unsupported claims and weak source material.
Turn missing visibility into an editorial question
If competitors are cited for a comparison you should be able to answer, inspect the cited pages. What do they explain? Which evidence makes them useful? Is your own page missing the answer, or does it bury it under promotional copy?
This is where the audit should earn its keep. If your comparison page avoids the difficult trade-offs while a cited competitor explains them clearly, there is an editorial job to do. A useful comparison, documented method or original example gives the reader something better. Repeating the prompt in more headings does not address the missing explanation.
Measure changes alongside conventional search performance. AI visibility is an additional view of discovery, not a reason to abandon the data you can already interpret.
A small audit log with honest denominators
Record one row per prompt and run. Keep the exact question, question group, platform, visible mode, date, session conditions, answer, cited URLs, brand mention, factual errors and proposed action. Do not combine a brand mentioned without a link with a citation to your own page.
| Measure | Illustrative calculation | Boundary |
|---|---|---|
| Observed mentions | 8 answers mentioning the brand ÷ 30 recorded answers | 26.7% of this sample, not market visibility |
| Observed citations | 3 answers linking to your site ÷ 30 answers | 10% of this sample; not a referral count |
| Accuracy problems | 2 answers with a documented material error | Record the wrong statement and correction source |
| Referral visits | Measured separately in analytics | Attribution and tracking limitations still apply |
These invented figures illustrate reporting only. If the next review adds ten easier branded questions, do not compare the resulting overall percentage directly with the original set. Report the unchanged questions separately and label new coverage.
Prioritise an incorrect service description, an inaccessible useful page or a missing comparison before chasing small changes in mention rate. Assign each action to a page or public source and record the change date. Use the SEO measurement guide alongside this log, and use the measurement plan to establish what business decision the audit informs. Repeated observations can support an investigation; they do not prove your edit caused a different generated answer.
Report what the audit can support
Show the question set, sample size and observed changes. Highlight incorrect descriptions that need repair and useful topics where the site lacks a credible answer.
Our test is simple: did the audit change a worthwhile publishing decision? Correcting a misleading description or answering a neglected buyer question is a defensible outcome. An impressive visibility percentage, with no clear sample behind it and no consequence for the work, is much harder to justify.




