How can you measure GEO and AI search visibility?

What does it mean to measure GEO and AI search visibility?

Measuring GEO means regularly checking when a business name or website appears in AI-powered search, whether it receives source citations, whether visitors arrive from those sources and whether this leads to meaningful business enquiries. It requires several different data sources: there is no single 'GEO ranking' that applies to every AI search engine.

In traditional SEO, a specific search term and ranking position often make a useful starting point. With AI-generated answers, the wording of the question, source material, interface and user context may all change. That's why recording that a chatbot named a company first once is not enough.

What does each AI visibility metric actually show?

A useful report does not mix up an AI system's answer, Google's impressions, Bing's citation counts and your own quote requests.

Google Search Console: how can you check AI appearances?

In 2026 Google introduced separate Generative AI performance reports in Search Console, including visibility data for Search AI Overviews and AI Mode. These reports show impression data for URLs appearing in generative features, not a stable conventional 'AI ranking' tied to a specific keyword.

1. Select the correct website property. Our guide to setting up Google Search Console explains the basics separately. Here we assume ownership has already been verified.
2. Open the generative AI performance report if it is available in your account. The exact menu location can change with interface updates, so don't rely entirely on an older screenshot.
3. Choose a comparable time period. For example, compare the last complete 28 days with the preceding 28 and record any significant site changes during that time.
4. Review pages and countries. Available breakdowns can show which URLs received AI appearances and in which countries. A device breakdown may also be available for Search.
5. Don't confuse counts from different performance reports. General Search Performance and AI-specific views may aggregate data differently. The separate view is supplementary, not an identical copy.

Google's official announcement lists views for impressions, pages, countries, devices and dates. This is not a report on ChatGPT appearances.

Bing Webmaster Tools AI Performance: what do citation counts mean?

The Microsoft Bing Webmaster Tools AI Performance view reports source citations in Bing and certain Microsoft AI experiences. It works differently from Google's reporting of AI appearances.

1. Open your website in Bing Webmaster Tools and find the AI Performance report.
2. Review the Total Citations figure: this counts source citations displayed in AI answers on supported interfaces, not clicks.
3. Review Pages Cited / the page-level breakdown: which URLs actually appear as sources?
4. Read the Grounding queries: these are sampled search terms used by the system, not necessarily the user's full original question.
5. Compare periods: is there a sustained change, and which pages does it affect?

For example, five Bing citations for a Knowledge Base article mean it was cited five times in supported AI interfaces during the period. It does not mean five customers arrived or the article 'ranked first' five times.

The official Bing AI Performance announcement provides detailed definitions.

What is Bing Topics Portfolio, and when does it help?

In October 2026 Bing also announced Topics Portfolio (beta) within AI Performance. It groups similar grounding queries into topics and may indicate where a site appears, which topics receive many citations and where competition is visible.

It isn't available for every website: Bing requires sufficient data. Topic size and opportunity estimates are indicative, not exact keyword search volumes or counts of paying customers. Topic classifications can also be wrong.

If it is available in your account, for example, you could compare the citation patterns for 'website redesign' and 'CRM integration'. These are illustrative topics, not real account data.

Source: Bing Topics Portfolio – official explanation (2026).

How can you test AI search manually without introducing bias?

Manual tests in ChatGPT, Google AI Mode or Copilot are useful observations, not representative measurements. Answers can depend on the account, language, query wording, region and service version. Don't keep rewording a question until your business finally appears once.

1. Record 8–12 genuine customer questions. Include choosing a provider, price and feature comparisons, and a specific problem. Don't arbitrarily change the question set midway through the measurement period.
2. Record the environment. Date, platform, search interface, country, language and account type if known.
3. Repeat using the same protocol. For example, run two or three control checks, but don't call the results a national share of visibility.
4. Distinguish mentions from links. Did the company name appear? The exact domain? Was there a clickable source link?
5. Keep a change log. Record page update dates, topics and the limitations of each observation.

Illustrative example: a provider tests the question 'Which Hungarian specialist should I choose for…?' each week. One favourable response does not prove first place nationwide. Several repeated observations under consistent conditions, combined with available platform reports, offer a more interpretable picture.

GA4: the business impact of leads from AI

A citation in an AI answer is not a customer. Business owners need to know whether visitors reach the right service, request quotations and generate valuable enquiries. GA4 can help track measurable outcomes, but not all traffic from AI search can always be identified accurately.

Google's Analytics traffic acquisition help provides further general information about measurable source data.

Example GEO measurement worksheet and 30-day routine

The following is a fictional methodological example, not actual Kaluzsa results or an export that can be generated by a single tool. Information from several systems must be combined, often with cautious interpretation.

1. Days 1–3: record the baseline and pages to measure.
2. Days 4–10: prepare a fixed control-question set and separate measurement platforms.
3. Days 11–24: record observations and available AI reports each week using the same method.
4. Days 25–30: compare signals with real website changes and lead quality. Lack of data is itself a finding; don't manufacture conclusions.

These 30 days are a measurement routine, not a promise that GEO visibility will improve in that time.

What mustn't you claim based on a GEO report?

• 'We're number one in Hungary' – one or a few AI answers do not constitute a permanent national ranking.
• 'Bing brought us 30 customers' – 30 citations do not mean 30 clicks or 30 customers.
• 'Google recommends us for every question' – the GSC AI report is not a complete query-by-query database of named recommendations.
• 'We have no AI report, so we're invisible' – data or permission limitations may be responsible.
• 'One graph rose, therefore our new article definitely caused it' – other search and technical factors may also influence changes.

A smaller, clearly labelled observation system is better than a spectacular but misleading 'AI rankings' table.

Frequently asked questions about measuring GEO

Related Knowledge Base guides and official sources

To understand measurement, review basic Search Console concepts first, then the generative search interfaces.