How to measure AI search visibility without a paid tracker

Measure AI search visibility by repeating a defined set of customer questions and recording the brands and source URLs each assistant returns. Keep those observations separate from technical readiness scores, connected-tool tests and visits to your website.

You can start with a spreadsheet. The useful work is deciding which questions represent customers, keeping the test conditions clear and retaining the evidence. A percentage without its questions and observation count is difficult to act on.

Which questions should I test?

Choose a small set of decisions that could lead someone to your business. Include a question about the problem, a question comparing solutions and a question about a practical next step. Use wording a customer could plausibly type.

For a UK accountancy firm, illustrative questions might concern moving bookkeeping systems, preparing records for a first accountant or comparing ways to organise invoices. These are example research prompts, rather than observations that the firm appears in any assistant.

Keep a fixed core set for comparison and a separate list for new questions. If the core list changes every week, a movement in the result may reflect a different test. Record a country or town only when local relevance matters to the customer decision.

What information belongs in each observation?

Save enough detail for someone else to understand the result. Avoid recording customer personal data or internal account details in a public report.

Use separate values for absent, unavailable and unknown. An assistant naming a business without a link is a mention. A linked source is a citation. A request made to your API is tool execution. None of those, by itself, establishes that a customer visited or contacted you.

How do I keep the test reasonably fair?

Start with a fresh conversation for the organic discovery test. Do not provide your URL, attach your brochure or connect your own MCP server before asking whether the assistant discovers you. Those actions change the information available to it.

Run connected-tool tests separately. They are useful for checking whether an assistant can use a service after it has access, but they answer a different question. The AEO Lookup API and MCP documentation explains that integration path.

Use comparable questions and save every result, including the ones where the business is absent. A single answer can be interesting; a repeated pattern is more useful. If you change a page, note the date and allow time for relevant discovery systems to revisit it.

What can Search Console and analytics tell me?

Google documents AI-feature traffic within Search Console's Web performance reporting. Do not label the entire Web total as AI traffic or invent a separate report that your account does not provide.

Browser analytics can show visits and referrers that the installed service actually records. Some visits arrive without a useful referrer. Keep attribution uncertain in those cases, and check the analytics provider's definitions before combining numbers.

For a small UK firm, the outcome worth following could be an enquiry, a booking or an accepted quote. Record the source evidence available for that outcome. A rise in citations with no measured change in enquiries remains useful evidence, with the commercial result recorded separately.

How should I summarise a week?

Publish the observation count with any rate. An illustrative calculation of 3 cited answers out of 12 comparable observations is 25%. It describes that small set of observations; it is not a probability that every future customer will see the business.

Keep one row for each target question, then a short list of changes and next actions. If no analytics service was running, record visitor totals as unknown. If a product did not offer search in the account being tested, record that condition rather than treating the result as a failed search.

Before investigating absent citations, follow the website access checks. Use the free AEO checks to diagnose observable technical issues. The AEO, SEO and GEO guide explains how those checks fit beside the observation log.

Which questions come up most often?

Is an AEO score my AI search ranking?

It is a score against the checks that tool defines. A ranking or citation observation needs its own question, assistant, date and evidence.

Can I test with my own tool already connected?

Yes. Record it as a connected-tool test. Use a separate fresh conversation for an organic discovery observation.

How often should I repeat the questions?

A weekly core set is a practical starting point for a small business. Retain the conditions and observation count, and review the cadence when enough comparable evidence exists.