A GEO audit checks how your business appears in AI-generated answers, what information supports those answers and which website improvements deserve attention. A useful audit produces evidence you can review and a short list of actions.

For an Australian business, that means testing questions relevant to your services and actual market. Appearing in an answer about a different country, customer type or service tells you little about your ability to attract suitable enquiries.

This GEO audit checklist explains how to build a repeatable test, separate different kinds of visibility and connect the findings to practical website work.

1 Choose the services and locations that matter

GEO stands for generative engine optimisation. In an audit, the aim is to understand visibility within generated answers, including business mentions, recommendations and citations linking to source pages.

Begin with one commercially important service and its relevant market. Write down what a suitable customer needs, where you genuinely operate and which competitors a buyer might reasonably compare.

A Sydney renovation company might focus on bathroom projects on the Northern Beaches. A consultancy delivering work remotely might use an Australia-wide market instead. Those businesses need different questions and different evidence.

Use actual customer enquiries to choose the topic. The audit should inform a business decision, such as improving a service page or explaining an offering more clearly. Avoid choosing questions simply because they are likely to produce a flattering answer.

2 Build a small set of realistic customer questions

Choose around ten questions for your starting sample. Include several buying stages so you can see whether visibility changes between learning about a service and choosing a provider.

The examples below illustrate how a hypothetical renovation company could organise its questions. Adapt them to your service and use locations you actually serve.

These prompts are a research sample. They are not evidence that a certain number of Australians search each phrase.

Keep branded questions in a separate group. Asking “What services does Example Bathrooms offer?” checks how the business is represented when it is already known. It does not show that the platform would introduce the business to a new customer.

3 Record the testing conditions

Use the same questions when you repeat the audit. Record the platform, date, exact prompt, Australian location context and whether web search was used. Note the model or mode when it is visible.

Start a fresh conversation for each test so earlier messages do not influence the question. This reduces one source of variation, although it does not remove every effect of account settings, personalisation or platform changes.

Save the full response and its visible source links. A screenshot can help preserve the layout, while copied text makes later comparison easier.

For a manageable baseline, test ten questions three times across a week on each platform you select. That gives thirty attempts per platform. Start with the platforms relevant to your customers and expand when you have a reason to do so.

Record failures separately. A tool error, refusal or unavailable answer is different from a valid answer that omits your business.

4 Separate mentions from recommendations and citations

These observations answer different questions and should have separate fields in your audit log.

A response can include several of these. Your name might appear because a directory lists you, while none of your own pages is cited. Another answer might cite your article without recommending your services.

Check that the name refers to the correct business. Similar trading names can create misleading results, particularly when the answer does not provide a location or website.

For branded tests, compare the response with the facts you publish. Log incorrect service areas, obsolete contact details and unsupported claims individually so you can investigate them.

5 Report the sample clearly

Use counts and denominators so another person can understand the result.

Suppose one platform produces thirty valid AI answers in a hypothetical test. Your business is mentioned in six, recommended in three and your website is cited in two. That means a 20% mention rate, a 10% recommendation rate and a 6.7% website citation rate within that test sample.

Those figures can overlap. They do not describe your share of Australian searches or prove how many people saw your business.

Keep results separate by platform and question type. Provider-discovery questions deserve particular attention when the goal is new enquiries. A combined score can hide strong branded results and weak discovery results.

For Google AI Overviews, also record whether an AI Overview appeared. If eight of ten searches produce one, report that coverage alongside mentions within the eight generated answers. This preserves the distinction between an absent feature and an answer that names other businesses.

6 Inspect the sources behind relevant answers

Open the sources cited in answers that mention you or suitable competitors. Identify the specific information those pages provide.

A competitor's service page might explain project scope clearly. A directory might supply a category and location. A project article might demonstrate experience relevant to the question.

Ask what a customer can learn from each source and whether your own information answers the same concern adequately. Useful findings include missing service details, unclear locations or an absence of relevant project examples.

Avoid drawing a technical diagnosis from an omission alone. If your business is absent, the answer does not reveal every source considered or why each was excluded. Investigate access, relevance and evidence separately.

When an answer gives incorrect information, check any visible source before changing the website. The error may come from an outdated profile you control, an external page or the generated response itself.

7 Check the website foundations

Review the homepage and the service pages relevant to the audit. Check access restrictions, accidental noindex settings, broken URLs and the links that help visitors reach important information.

For Google's AI features, its official documentation says a supporting page must be indexed and eligible to appear in Search with a snippet. Use Search Console to investigate Google's access and indexing separately from your prompt tests.

For ChatGPT search, OpenAI documents OAI-SearchBot and GPTBot separately. Search access and model-training permissions serve different purposes; review the search bot and hosting restrictions relevant to discovery rather than assuming every AI bot needs access.

Then review the content itself. Can a prospective customer establish what you do, where you operate, what the service includes and why you are suitable? Are those facts consistent with the public profiles you maintain?

Keep findings specific. “The page does not explain which project types are accepted” gives your team a clearer task than “improve AI readiness”.

8 Use platform reporting alongside prompt tests

Prompt tests provide observations about selected questions. Platform reporting and website analytics provide different evidence.

Google says traffic from AI Overviews and AI Mode is included in Search Console's overall Web performance reporting. Those totals should not be presented as a separate measurement of traffic from AI answers.

Microsoft's AI Performance reporting in Bing Webmaster Tools provides citation information across supported Microsoft and partner experiences. Its June 2026 preview update adds context including intents, topics and citation share. That coverage does not represent every AI platform.

In your analytics, review identifiable AI referrals and whether those visitors enquire. Referral data can be incomplete, so also record how prospects say they found you. Treat those accounts as useful evidence while recognising that attribution is imperfect.

A business mention, a website visit and a qualified enquiry are separate outcomes. Reporting them separately makes the audit more useful.

9 Turn findings into a short action plan

Prioritise findings by the problem they create and the evidence available. A confirmed access restriction on a key page deserves a different response from one disappointing prompt result.

For the first month, a practical order is:

1. Resolve confirmed access problems. Check that intended public pages can be reached by the relevant search systems.

2. Correct business information. Update inaccurate service, location and contact details on pages and profiles you control.

3. Improve the priority service page. Answer unresolved buying questions and add relevant, verifiable project evidence.

4. Repeat the original tests. Keep conditions as consistent as possible and record the changes made between rounds.

Assign each task to someone and identify how completion will be checked. “Publish a project example showing a small bathroom renovation” is reviewable. “Become more authoritative” leaves the work undefined.

If results improve, preserve the evidence and continue monitoring. Without a controlled experiment, a change in responses alone does not prove which website edit caused it.

What is a good GEO audit score

There is no universal pass mark for every business and platform. Assess your relevant discovery questions, factual accuracy and commercial outcomes against your own baseline. A score without its prompts, testing conditions and calculation is difficult to interpret.

How often should an Australian business repeat a GEO audit

A monthly review is a practical starting point for a small business. Repeat the relevant checks after significant service, location or website changes. Consistent testing is more informative than frequently changing the questions to obtain a better result.

Creato helps businesses improve their websites and content structure. If your audit reveals unclear service pages or a website that needs rebuilding, send us your URL and project details.