“Are we visible in AI answers?” Measure the evidence, not the excitement

A dashboard gives the company an “AI visibility score” of 72 out of 100. Another tool finds the brand in three of twenty prompts. Someone tests ChatGPT manually and sees no mention at all.

Which number is correct?

Possibly all three, because they are measuring different samples, products and moments. The mistake is treating them as if they describe one objective market share.

Google made the measurement problem materially better in June 2026 when it announced dedicated generative-AI performance reports in Search Console. The reports are designed to show impressions and dimensions such as pages, countries, devices and dates for supported generative features including AI Overviews and AI Mode.

There is an important publication-date caveat: on 28 August 2026, these reports were still rolling out to a subset of sites. Do not write the historical article as if every property already had access.

Measure AI Search visibility with an evidence hierarchy

Instead of one synthetic number, use several layers and label them correctly.

Google announced dedicated generative-AI performance reporting in Search Console on 3 June 2026. The initial rollout was limited to a subset of websites.

Author recommendation: label third-party prompt trackers as sampling systems. They can be very useful without pretending to be census data.

An AI impression is real data, but it is not a business outcome

If a service page receives more impressions inside a supported Google generative feature, that tells you something concrete: Google showed the URL in that environment.

It does not automatically tell you whether the user noticed the brand, whether the page was a primary grounding source, whether the answer described the company accurately or whether the eventual phone call came from that exposure.

For a small business, build a simple chain:

AI visibility → visit → qualified action → CRM outcome.

Some links will remain fuzzy. Google itself documents normal differences between Search Console and Analytics. A user may read an AI answer, remember the company name and call directly the following week. Your systems may never connect that journey.

Do not “solve” the gap by assigning a made-up conversion value to every impression. Put uncertain effects in a separate category.

Search Console cannot measure every assistant

The Google report is valuable because it is first-party data. It does not tell you everything about ChatGPT, Copilot or other products.

For those systems, a repeatable prompt set is still useful. Build a small benchmark around actual customer research tasks rather than vanity prompts. A Swiss B2B firm might create 30 cases across German, French, Italian and English: local discovery, service comparison, technical questions and commercial intent.

Run the same categories on a regular schedule and record the product/model context. Measure more than whether the brand name appears. Track whether a source is linked, whether the facts are correct, whether competitors are present and whether the recommendation actually fits the query.

This produces a trend line. It does not produce a universal percentage of all AI conversations.

That distinction should appear directly in the report so management knows what the KPI means.

Establish the baseline before changing the site

The fastest way to make AI SEO impossible to evaluate is to change ten things at once and start measuring afterwards.

Before a major content project, save the available Search Console baseline. Segment by page type, language and country. If the dedicated GenAI report is not yet available for your property on 28 August, record that fact and use the normal search data plus your controlled prompt sample until access arrives.

At the same time, note meaningful interventions: new service pages, major rewrites, structured-data changes, PR coverage, social campaigns and technical releases.

When the metrics move, you can then say, “visibility rose after these changes” rather than “this one sentence caused a 27% AI ranking increase”. The first is a defensible observation. The second usually needs evidence you do not have.

Multilingual Swiss sites need extra discipline. A German page and a French page can have very different search demand and AI exposure. Combining them into one “AI score” can hide both improvement and decline.

A minimum measurement pilot for an SME

Choose one commercially important service cluster and run the following process.

1. Check whether the property has access to Google’s generative-AI performance report on the publication date.
2. Export available impressions by page, country and date.
3. Map those landing pages to analytics events and qualified CRM leads.
4. Create a fixed 20–30 prompt sample for assistants not covered by first-party data.
5. Record content changes and compare sufficiently long pre/post periods.
6. Add a plain-English definition to every KPI: “proves X; does not prove Y.”

Useful metrics are GenAI impressions by commercial page, qualified visits to those pages, lead rate from the page cluster, citation rate in the fixed prompt sample, factual accuracy of AI mentions, and branded-query trend.

Brand demand is deliberately an indirect metric. If it rises, AI exposure may be involved; so may PR, social, offline marketing or existing customers.

The practical conclusion: AI Search is becoming more measurable, but it is not completely observable. Use official first-party measurements where they exist, controlled sampling where they do not, and never let a tidy score hide an untidy methodology.

One prompt result is not ranking evidence. Location, account and timing alter outputs. Fixed questions and a fixed review date still reveal trends. Our guide to Search Console platform properties adds social and video measurement.

Useful pages answer a concrete question, show experience and remain technically accessible. Our web design and SEO approach builds that foundation instead of selling a badge.

Sources

• Google Search Central — Generative AI performance reports
• Search Console Help — Generative AI performance report
• Google Search Central — Optimizing for generative AI
• Google Search Central — Search Console and Analytics

---