Measurement

How to Measure AI Search Visibility Without Misleading Metrics

By GEO Specialist Perth. Updated . 7 minute read.

AI search visibility is not one number. Learn how to run repeatable prompt checks, record mentions and citations, connect qualified referrals to enquiries, and avoid claiming causation that your data cannot prove.

Take Free AI Readiness Quiz Book My Free Strategy Session

AI search visibility is not a single ranking position. For a Perth service business, a useful measurement system separates what an AI answer says, whether it links to you, whether anyone visits, and whether a genuine enquiry follows. That separation prevents a reassuring citation count, or a sudden traffic change, being mistaken for proof that GEO work caused revenue.

Google’s newer Search Console documentation provides a Generative AI performance report for impressions in AI Overviews and AI Mode. Its data is also included in the broader Web search performance data. The AI report is an impressions view, not a separate measure of qualified leads or all AI-platform referrals. [1] Microsoft’s Bing Webmaster Tools AI Performance public preview provides a different view: citations, cited pages, grounding queries and trends across supported Microsoft AI experiences [2]. These tools are useful, but they answer different questions.

Start with a dated measurement baseline

Before changing a page, save a baseline dated with the exact start and end dates. Record:

  • the pages and services you want to monitor
  • the locations and customer problems represented by your prompts
  • existing organic clicks, impressions, click-through rate and conversions
  • direct enquiries, phone calls and booked consultations where they can be identified
  • the AI products and access conditions tested

Keep the baseline in a simple spreadsheet or reporting document. Note the device, location setting, logged-in state, language, date and exact prompt. AI responses can vary, so a result captured without those details is difficult to reproduce. Do not treat the baseline as a promise of future performance. It is a dated comparison point.

Test prompts consistently, not selectively

Create a small prompt set that reflects real buying journeys, not just your business name. For example, test questions about choosing a service, comparing approaches, checking typical project considerations, and finding a Perth provider for a clearly described need. Include relevant suburbs only where location genuinely changes the answer.

Use the same prompt wording and settings at each review. Run the set on a fixed cadence, such as monthly, and record the complete answer or a concise transcription of the relevant passage. Log:

  1. whether your business is mentioned
  2. whether your website is linked
  3. which page is linked or cited
  4. the wording used to describe the business
  5. whether the information is correct
  6. competitors or other sources shown
  7. the date and platform

A single successful test is an observation, not a ranking. Repeated observations across the same defined prompt set can show a directional pattern, but they do not represent every user, prompt, model or location.

Separate four visibility signals

Mentions

A mention means the business or brand appears in an answer. It may be unlinked, incomplete or inaccurate. A mention can indicate that the system associated the business with a topic, but it does not show that a potential customer saw it or trusted it.

Citations and linked sources

A citation is a source displayed or linked as support for an answer. Bing’s AI Performance documentation defines total citations as the number of citations displayed as sources in AI-generated answers during the selected period. It specifically cautions that the measure does not indicate placement or presentation in an individual answer [2]. Treat cited pages as a source visibility signal, not a conventional position or authority score.

Referrals

A referral is a visit that reaches your site from a platform. Use analytics with suitable campaign tagging where you control the link, and inspect referral and landing page data. Google Search Console can show clicks from Google Search, but its documented click metric is a click that sends someone outside Google Search, Discover or News, with result type details affecting methodology [3]. Do not assume every AI answer visit will be labelled consistently, and do not infer an AI source from an unattributed direct visit.

Enquiries and qualified outcomes

Track the actions that matter to the business: a completed contact form, a call, a booked strategy session or a qualified project conversation. Ask how the person found the business, while recognising that recall is imperfect. Keep source data, self-reported attribution and assisted discovery separate. A person may read an AI answer, search the business later and then contact you through another channel.

Use each platform for the question it can answer

Search Console is appropriate for monitoring Google Search performance trends. Its impressions mean a link was seen or potentially seen, clicks measure interactions that lead away from Google, and average position is a calculated Search metric, not a universal AI answer rank [3]. Compare consistent date ranges and relevant page or query groups. Avoid describing a change in Web search clicks as AI-generated traffic unless the available data identifies that source.

For Google AI visibility specifically, use the Generative AI performance report to compare impression trends by page, country, date and device. Google says the insights were rolled out worldwide on 31 August 2026, but a property with insufficient AI impressions may not show the report. It covers AI Overviews and AI Mode, excludes Search Labs experiments, and does not establish that an impression became a visit or lead. Avoid adding its totals to the broader Web report as though they were separate audiences. [1]

Bing Webmaster Tools AI Performance is useful for supported Microsoft surfaces, including the experiences named in its announcement. Review total citations, average cited pages, grounding query samples and page level activity together. Bing states that grounding queries are a sample and that cited page activity reflects how often pages are cited, not page importance, ranking or placement [2]. Public preview data may evolve, so record the report version or access date in internal notes.

Analytics and your customer relationship process complete the picture. Use landing pages, referral data, form submissions, call tracking where lawful and practical, and enquiry notes. These are business measurement practices, not proof that an AI platform caused a conversion.

A practical Perth example

Hypothetical example: a Perth building consultancy tests eight prompts about selecting advice for a renovation project. In April, the business is mentioned twice, linked once, receives no identifiable referral, and records three enquiries from unknown sources. In May, it is cited three times, receives two visits from a tagged link, and records four enquiries, one of which says they found the business through an AI answer.

The defensible report says visibility and observed enquiries increased in May, with one self-reported AI-influenced enquiry. It does not say the extra enquiry was caused by the additional citation. Other marketing, seasonality, brand awareness and normal variation remain possible explanations. The next action is to continue the same test and improve the accuracy and usefulness of the cited service page, rather than celebrate a causal win.

Use a monthly record with columns for test date, platform, prompt, mention, citation, URL, referral, conversion and notes. Show counts and examples, not a blended score that hides what happened. Annotate site changes, campaign activity and tracking changes. Compare like with like, and avoid small samples presented as a market share.

For Google AI features, ordinary SEO foundations remain relevant. Google’s current guide also requires the site to be included through its Search generative AI control, while special AI schema is not required. Eligibility and indexing still do not guarantee inclusion. [4] Measurement should therefore evaluate discoverability and business usefulness, not promise an appearance.

For related foundations, see what generative engine optimisation means, SEO versus GEO, and our GEO SEO retainer. If you need a clearer measurement plan for your service business, take the Free AI Readiness Quiz or Book My Free Strategy Session.

FAQs

Is a mention the same as a citation?

No. A mention names the business, while a citation identifies a source, usually with a link or displayed reference. Either can be inaccurate, and neither proves a visit or enquiry.

Can Search Console isolate AI Overview traffic?

Yes, Google now provides a dedicated Generative AI performance report showing impressions for AI Overviews and AI Mode when the property has sufficient data. Its data is also part of broader Web performance. It does not isolate AI-generated enquiries or provide every metric in the regular performance report. [1]

How often should prompts be tested?

Use a cadence you can maintain, commonly monthly for a small service business. Consistency, exact prompt records and dated notes matter more than testing many prompts once.

What is the most meaningful AI visibility metric?

There is no universal winner. For management decisions, connect repeated visibility observations with qualified referrals and enquiries, while keeping each signal labelled and acknowledging attribution limits.

Sources

  1. Generative AI performance report Search, Google Search Console Help
  2. Introducing AI Performance in Bing Webmaster Tools Public Preview, Microsoft Bing
  3. What are impressions, position, and clicks?, Google Search Console Help
  4. Optimizing your website for generative AI features on Google Search

Related GEO and SEO guides