The Search Console Generative AI Report: Measuring Visibility Without Clicks

What Actually Shipped

For roughly two years, the honest answer to "how much of our organic performance comes from AI Overviews?" was that nobody could tell you. Google counted AI Overview and AI Mode impressions and clicks inside the standard "Web" search type, aggregated with classic blue-link results and impossible to separate. Every attempt to size the AI surface relied on third-party SERP scraping or inference from traffic decline.

That changed in June 2026. Google announced generative AI performance reporting in Search Console — a dedicated view of how often a site's URLs appeared inside Google's generative AI features. It launched to a limited set of properties, driven in part by regulatory commitments in the UK, and widened over the following months.

The report is genuinely new information. It is also far more limited than most teams assume from the announcement, and the limitations determine what you are allowed to conclude from it.

What Is In the Report, and What Is Not

The distinction between present and absent fields is the entire analytical story. Everything you can legitimately do with this report follows from the right-hand column.

Google has stated it may add metrics over time based on site-owner feedback. Build your reporting so that adding a clicks column later is an enhancement, not a rebuild.

The Reconciliation Trap

The first instinct of most analysts seeing two reports is to difference them. Total Web impressions minus generative AI impressions should equal classic organic impressions, and from there you should be able to isolate what the blue links are still doing. This is wrong, and it produces confidently incorrect board slides.

Generative AI impressions are a subset of the standard Web totals, not a parallel channel sitting alongside them. The subtraction does produce a residual impression count. But AI-surface clicks are also inside the Web click total and cannot be extracted, so the residual impressions have no matching click figure. Pairing "classic impressions" with "all Web clicks" yields a CTR that is structurally overstated — and the overstatement grows precisely as AI surfaces take more of your query mix, which is the opposite of what you want from a diagnostic.

The correct mental model: the standard report tells you what the whole Search surface did. The generative AI report tells you how much of the impression volume came from the AI layer. There is no third view that separates the click behaviour, and constructing one from arithmetic invents data.

The Visibility Ledger

AI-surface visibility is now a four-source measurement problem, and no single source answers more than a quarter of it. We use a Visibility Ledger to keep the sources in their lanes — each row states the one question that source can answer honestly, and the question it will be misused to answer if you let it.

Rows one and two are Google-only. Rows three and four are where third-party assistants become visible, and they require their own instrumentation — the four-layer approach in our guide to AI search attribution covers the build. A team that has only the new Search Console report has one row of four and should present it that way.

Two Metrics Worth Computing

Impression-only data supports exactly two durable metrics. Both are trend instruments — there is no benchmark to hit, only a direction to watch against a fixed page set.

Neither metric tells you about money. Any claim connecting AI visibility to pipeline still has to route through channel-level economics — the reference ranges in our 2026 CAC benchmarks and the ceiling logic in the LTV:CAC framework apply here exactly as they do to any other acquisition surface.

Reading the Four Patterns

With impressions on one side and total Search performance on the other, four combinations are diagnostically distinct. Each implies a different response.

Patterns two and four both demand a volatility check before action. The four-step method in our guide to telling a real algorithm hit from a reporting artefact applies unchanged, and the short history available in the generative AI report makes that discipline more important, not less — you do not yet have a year of seasonality to compare against.

The Opt-Out Control

Alongside the report, Google introduced a control letting sites exclude their content from generative AI features, and stated that using it does not affect core Search ranking. The control exists largely because of regulatory pressure on how publisher content is used in AI answers.

For a commercial business, the arithmetic is unambiguous. Opting out means no impressions and no traffic from AI Overviews, AI Mode, or generative Discover. You are choosing absence from the surface where an increasing share of early-stage research now happens, in exchange for protecting click volume on a surface you have already partly lost. That trade makes sense for a publisher whose revenue is per-pageview. It rarely makes sense for a SaaS platform, a DTC brand, or a professional-services firm whose revenue depends on being in the consideration set.

This is a different decision from crawler policy, and the two get conflated constantly. Whether Google may use your content in AI answers is a visibility question. Which AI bots may access your site at all — for training, for retrieval, for agent browsing — is a separate access question with different economics on each bot class. We cover that separately in which AI crawlers to allow, block, or charge.

A 30-Day Instrumentation Sequence

The report rewards patience. The first month is baseline construction, not optimisation — there is no click signal to optimise against, and acting on four weeks of impression data is how teams talk themselves into rewriting pages that were performing.

Where This Sits in the Wider Programme

Measurement does not create visibility. The generative AI report tells you whether the work is landing; the work itself is content structured to be retrieved, quoted, and attributed — the five-lever approach in our guide to generative engine optimization. Read together, they form a loop: GEO produces the citable asset, the AI report confirms Google is drawing on it, and the attribution stack establishes whether any of it reached revenue.

For UAE and GCC businesses the country dimension is more useful than it first appears. AI surface triggering, language mix, and competitive density differ sharply between Gulf and Western markets, and a property reporting healthy aggregate impressions can be effectively absent in its actual target geography. Splitting by country before drawing conclusions is standard practice in our Dubai and GCC engagements.

The broader operating model — how measurement, content, and paid decisions connect rather than run as separate workstreams — is set out in the Kres Labs Growth Playbook.

Frequently Asked Questions