An AI visibility audit should leave you with something more useful than a screenshot of your business in an answer. It should show which customer questions you tested, what appeared, whether it was accurate and what you will improve next.

This worksheet helps an Australian business run that process consistently. Use it for an AEO audit, a GEO audit or an internal review of how answer platforms describe your services. The labels vary across the industry; define the platforms and questions in your own report rather than assuming the label describes a standard test.

The result is a documented sample of visibility, not a ranking certificate or a forecast of enquiries. For the wider content strategy behind this work, read Creato’s SEO for AI guide: SEO for AI guide.

A repeatable audit measures a defined sample, not a guaranteed ranking. Original Creato editorial graphic.

View and share the full-size infographic

1 Define the decision before opening an AI tool

Choose a service you want suitable customers to find. Record its genuine service area, intended customer and main website page. A business serving one region should not judge itself against an Australia-wide question unless national discovery is commercially relevant.

Write one decision the audit will inform. For example: “Decide which information our commercial fitout page needs before its next revision.” That gives the findings somewhere to go.

Choose the answer surfaces separately. Google AI Overviews, Google AI Mode and a search-enabled chatbot are different test environments. Do not combine their results simply because they all generate text.

2 Build and freeze your question set

Draft questions from sales calls, enquiry forms and concerns customers raise before buying. Keep the initial set small enough to repeat carefully. Ten prompts is a practical starting point, not a statistically representative sample.

For a hypothetical Australian fitout company, useful prompt patterns might include:

• Provider discovery: Which companies handle small office fitouts in [actual service area]?

• Suitability: What should a tenant check before choosing an office fitout contractor?

• Cost research: What affects the cost of fitting out a small office in [city]?

• Comparison: How does a design-and-construct fitout differ from hiring designers separately?

• Branded accuracy: What services does [business name and website] provide?

Replace placeholders before testing. Keep branded questions in their own group: supplying your name tests recognition and factual accuracy, not independent discovery.

Save the exact wording and a prompt ID. If you improve a question later, issue a new version instead of silently replacing it. Otherwise, the next result may reflect the new wording rather than a change in your visibility.

3 Record conditions and retain the answer

For each attempt, record date, time zone, platform, visible model or mode, search setting, account context and location context. If a model name is hidden, write “not displayed”. Distinguish a location you put in the prompt from the device or account location. They are not interchangeable.

Start each test in a fresh conversation where available. Avoid earlier messages that name your business. Record personalisation settings you can see; do not claim that a fresh chat removes every source of personalisation.

Repeat the fixed questions on separate occasions. Three runs per question across a week can provide a manageable baseline, but those observations are neither independent survey respondents nor a measure of market demand.

Save the full answer and visible links with the run ID. Record an error, unavailable feature or refusal as its own outcome. Keep unsuccessful attempts in the log instead of rerunning until you get a favourable answer.

4 Classify what actually happened

Use separate yes, no and unclear fields for these observations:

• Mention: the answer names the correct business.

• Recommendation: it presents the business as a suitable option, rather than merely discussing it.

• Owned citation: a visible source link points to your website.

• External citation: a visible source about your business points elsewhere.

• Accuracy issue: a checkable statement conflicts with confirmed business information.

Open citations and check the destination. A familiar name can refer to another company. A linked article can support one sentence without supporting every claim in the response.

For an accuracy issue, record the claim, the correct fact, the evidence and any source the answer supplied. Do not invent a source when none is visible. These records are more actionable than a single “AI score”.

5 Report counts with the right denominator

Report each platform and prompt group separately. Include the number of attempts, valid answers, feature absences and errors. Calculate mention and citation rates only after defining the denominator.

A hypothetical report might say: “Across 30 attempts, 27 produced valid answers and three failed. The business appeared in six of the 27 valid answers.” That is a 22.2% mention rate within this sample. It is not a share of all Australian AI searches.

For AI Overviews, also report how often an Overview appeared. “Eight Overviews from ten searches, with one brand mention” describes two observations. Hiding the two searches without an Overview makes comparisons harder.

Mentions, recommendations and citations can occur together, so do not add their percentages into a total visibility score. When no valid answers exist, report the rate as not applicable rather than zero.

6 Investigate before prescribing website changes

An omitted business does not reveal why it was omitted. Check access and content separately.

Check Google’s current AI optimisation guidance for eligibility, including indexing, snippet eligibility and the effective Search generative AI inclusion setting in Search Console. Record the reports and metric definitions actually available in the account. Do not treat all Search traffic as AI-attributed traffic or confuse AI impressions with visits or enquiries Google Search Central.

For ChatGPT search, OAI-SearchBot and GPTBot serve different purposes. Search discovery and model-training permissions can be managed independently. Check the relevant crawler and hosting restrictions rather than opening all bots indiscriminately OpenAI crawler documentation.

For the page-level checks behind this investigation, use our AEO website design guide: AEO website design guide.

Then inspect the customer-facing information. Does the service page explain accepted projects, location coverage, process and evidence of relevant work? Could a prospect verify the claims? If those answers are missing, improve the page for readers instead of repeating an AI keyword throughout it.

Creato’s website design work focuses on clear content and enquiry paths. Review the approach at website design approach when deciding whether a content correction or a broader website update is needed.

For an example of presenting a specialist business through branding and website design, explore our Michael Bell Architects project: Michael Bell Architects project. Treat it as a design reference, not evidence that a particular layout guarantees AI citations.

7 Turn observations into owned tasks

Each task needs a named owner, affected URL, supporting evidence, due date and completion check. “Explain the minimum project scope on the fitout page” is specific enough to review. “Improve authority” is not.

Prioritise confirmed business-information errors and access problems, then gaps in important service pages. Keep speculative ideas separate from established faults. Record what changed before repeating the original prompts.

Compare the next run with the baseline, but be cautious about causation. Platforms change, answers vary and several site updates may happen together. Better results after an edit do not prove that edit caused the improvement.

What to do with the worksheet

Copy the fields in the two-page audit worksheet into your own audit log and retain one record per attempt. Review trends alongside enquiries and identifiable referral visits. A brand mention, a website visit and a qualified lead are different outcomes.

Repeat when a business change gives you a reason to check again, or use a monthly review cadence you can maintain. Keep earlier runs so you can distinguish a recurring factual problem from an isolated answer.

If you want help turning the findings into a defined programme of work, explore Creato’s AEO and GEO services: AEO and GEO services. If the audit exposes unclear services or weak enquiry pages, talk to us about the website improvements worth making first: contact Creato. Bring the evidence log so the discussion starts with the actual problem.