Measuring Visibility in AI Search: A Practical Method

Google reports which pages appear in its own generative AI features through Search Console; that is, on its own, a verifiable data source. For other engines such as ChatGPT or Perplexity, an equally comprehensive, officially verified report is not generally available yet; visibility there gets observed largely by hand, by running queries.

That is why a single sentence like 'our AI visibility went up' is rarely precise enough; without naming which engine, which query and which date, that sentence is an impression, not a measurement.

This report breaks down a page's impressions in Google's generative AI features, AI Overviews and AI Mode, over time, by country and by device. The 'pages' dimension is based on the final URL a feature links to, after any redirects.

This data becomes meaningful once read alongside the classic search performance report: if a page's impressions are rising while clicks stay flat, the page is being seen but not chosen, which is usually a sign to revisit the title and description.

For engines Search Console does not cover, the most concrete method is building a fixed set of queries drawn from real customer questions.

• Write down fifteen to twenty questions customers actually ask, about price, service scope, location, and so on.
• Ask those same questions to ChatGPT, Perplexity and Google AI Mode within the same month, in the same order.
• Note for each result whether the brand's name appears and which page gets cited as the source.
• Log this consistently in a table: date, engine, query, result, source page.
• Repeat the same queries a month later and compare against the previous log, adding the change to the table.

Four basic fields need logging: the date, the engine used, the question asked, whether the brand appeared, and which page was cited as the source. Without these four fields a comparison is not really possible.

When comparing, the same engine needs comparing against itself and the same query against itself; folding a ChatGPT result and a Google AI Mode result into a single combined number is misleading, since the two run on different mechanisms.

A scene we saw at a real estate agency: the marketing team tried different engines with different questions every month, trying to keep results in memory; a year on there was a feeling of 'we appear more now' but nothing on record actually showed it. The decision criterion is this: an observation is not a measurement until it is logged in a table; deciding based on a feeling makes it impossible to trace, months later, which change actually worked.

A common one we have seen at a flooring manufacturer is running the query set once, concluding 'we appear', and never repeating it a month later to notice any actual change.

Another is focusing entirely on AI visibility and dropping classic Search Console data altogether; both are reflections of the same healthy site, one does not replace the other.

This level of measurement is not necessary for every business. For a small business offering a single service in a narrow area with limited competition, building a monthly query set can cost more effort than the insight it returns; a short manual check every three months is often enough there.

The decision criterion is how intense the competition is and how fast the brand is growing. A business expanding across several cities in a fast-moving sector gets real direction from regular measurement; a business serving the same customer base for years, with a stable set of competitors, has less reason to prioritise this investment.

If a business has not yet built basic visibility, Google Business Profile, consistent contact details, that is where the priority belongs first; measurement tracks how an existing visibility changes, it does not create a visibility that is not there yet.