Search Console and AI Overviews: how to read the generative AI report (and what it leaves out)

What the generative AI performance report is, and where it lives in Search Console

The generative AI performance report is the new section of Search Console (Performance > Generative AI) that shows how many times your pages showed up inside the answers Google assembles with AI. Google announced it on June 3, 2026 for a small set of sites and rolled it out to everyone on August 31, 2026. That is what most people mean when they search for Google Search Console AI overview — there is no separate panel with that name.

The report spans two surfaces with their own official names: AI Overviews — the block that appears above classic results — and AI Mode, the conversational tab that replaces classic search when a user turns it on. How AI Mode splits a single search into multiple sub-queries is its own mechanism with its own expansion logic, which we do not unpack here. Discover has its own separate generative AI report, and Search Labs experiments are not included in either.

Search Console, AI Overviews, and AI Mode: what the report shows and what it does not

This is where most early takes on the Search Console AI report get it wrong: they assume that, being a new Search Console report, it carries the same four metrics as regular Performance. It does not. It carries exactly one: impressions. Its dimensions are Pages (the final canonical URL, after any redirects), Countries (where the search originated), Devices (desktop, tablet, or mobile), and Dates — with hourly, daily, weekly, or monthly granularity, but in Pacific time, so the report's day boundary does not line up with midnight in Bogotá, Mexico City, or Buenos Aires.

Export works the same way it does everywhere else in Search Console, with one quirk worth knowing: when a cell has very low volume, the interface shows "~" or "-", and both turn into 0 once exported. If you are automating anything on top of this data, that detail skews averages unless you filter it out first.

The missing half — queries and position — is not an oversight on Google's part: it is exactly what you have to cross-reference Search Console with your daily rankings to fill in. The generative AI report tells you which of your pages showed up in an AI answer, in what country, and on what device; a keyword-level rank tracker tells you the exact query and organic position that same page is competing on.

How Google counts AI Overview impressions: one per site, one per block

This is the detail that trips up almost everyone the first time they open this report: the main chart and the per-page table almost never match, and that is not a bug. The chart aggregates at the property level: if two of your URLs show up inside the same AI Overview block, that AI Overview counts as a single impression in the chart, no matter how many of your links sit inside it. The per-page table, on the other hand, counts every URL separately.

The simplest example: you have 3 different URLs cited inside the same AI Overview for one search. The chart adds 1 impression for your property. The table, meanwhile, shows 3 impressions split across those three URLs. Neither number is wrong; they measure different things. The report also applies Search Console's standard impression rules: a link tucked behind a "Show more" inside an expandable section usually does not count as an impression until a user expands it.

The report inherits the same 1,000-row cap that already applies to general Performance, so on larger sites with thousands of cited pages, one export cannot show the full table — you need to segment by country or date range to see it all.

One note on Discover, even though its report is separate: counting there works differently and looks more like the classic feed — one impression per result and per user session, not per block.

Clicks, CTR, and position: the errors already circulating about this report

With such a new report, four mistakes are already spreading. Worth correcting each one before it becomes the explanation you give a client.

If what you actually care about is what happens to your real clicks once Google shows an AI-generated summary above your results — not just this report's impressions — that is a different problem, and we cover it in the guide on what happens to your clicks when an AI overview appears.

Country filter: reading the report when your site sells across several LATAM markets

The report's Country dimension records where the search originated, and it combines cleanly with the page filter. This matters for any business with a presence in more than one LATAM country, because AI features show up unevenly across countries and languages: Google can trigger an AI Overview for a query in Mexico and not for the same search in Argentina, or the reverse. The practical rule is to never compare countries against each other — each one should be compared against its own history, never against the site's total.

Because the report inherits both the 1,000-row cap and the 16-month retention window from general Performance, it is worth building a routine for exporting Search Console before it disappears. If you sell across several countries, freeze a baseline now: a weekly export by page × country, with a note on every content change you make. Without that baseline, six months from now you will not be able to tell a real shift apart from normal report fluctuation.

From the per-page report to your keywords: the crossover that completes the picture

The generative AI report hands you pages; your actual work happens at the keyword level. The bridge between the two is the "ranking URL": the URL on your site that shows up for each keyword you track. With that, you can build a simple crossover table like this:

An honest note before you expect more from this table: Rankiamos only covers the AI Overview row — whether it cites you, on the Pro plan, tied to Pro's Google AI overview citation tracking; AI Mode and Discover stay limited to whatever Search Console reports on its own, with no automatic crossover. The app does not import this report. That "cited?" column is built off the "ranking URL", not off a specific URL cited inside the AI Overview.

This table does not replace knowing who else gets cited alongside you on each query, or the full per-query tracking method — that lives in its own guide, with the method for seeing who else is cited on each query. The goal here is smaller and more urgent: knowing which of your pages is generating those AI impressions and which keyword to tie it to, before you waste time optimizing the wrong page.

To have queries, "ranking URL", and organic position in one place, the starting point is connecting your Search Console property to a rank tracker. Without that connection, the crossover table gets built by hand, page by page, and on a large catalog that stops being sustainable by week two.

The generative AI control in Search Console: what the switch does and how inheritance works

The Search Console generative AI control lives under Settings > Generative AI in Search, inside each property, and offers three options. Include is the default when there is no parent property above it. Opt out is the one that removes you from these features. Inherit from parent property is the default whenever a parent does exist — for example, if your property is a URL-prefix inside an already-verified domain.

The control spans all three surfaces: AI Overviews, AI Mode, and Discover's AI features. Choose Opt out, and according to Google Help three things happen at once: your links and content stop appearing in those features, your content stops being used to ground the answers Google generates, and you stop receiving impressions and traffic from those surfaces. Similar content from other sites keeps appearing normally, and in Google's own words, "it may look like yours." Google also clarifies three things the control does not do: it is not a ranking signal, it does not disable your presence in Merchant Center or Google Ads, and it does not affect model training — that runs through a separate mechanism, Google-Extended. The change applies within one to two days, with a bit more lag on pages Google has heavily cached.

Inheritance is the detail that catches people off guard the most. Say your domain property (yourstore.com) sits on Include, but you create a URL-prefix property for one folder (yourstore.com/blog/) and set that one to Opt out — that folder drops out of AI Search while the rest of the site keeps appearing. The reverse also holds: if someone with access to the domain property switches the general setting to Opt out, any child property left on "Inherit" gets opted out too, without you deciding that directly. Worth checking this setting even if you never touched it yourself.

Search Console control vs. nosnippet vs. Google-Extended vs. noindex: which one does what

Four mechanisms sound similar and do different things. The core message: the generative AI control is the only one that removes you from AI Search without touching your classic blue-link snippet.

The most common mistake is using Google-Extended thinking it removes your site from AI Overviews. It does not: per Google, Google-Extended only controls whether your content is used to train future Gemini models and ground answers inside Gemini apps and Vertex AI, and it does not affect your inclusion or ranking in Search. If your goal is to stop being cited inside an AI Overview, the right mechanism is the Search Console generative AI control, not robots.txt.

For the vast majority of LATAM sites — stores, local services, niche blogs, consultancies — opting out of generative AI does not fix a real problem and does close off a source of visibility. The core reason: when you opt out, the user's search still generates an AI Overview exactly the same, Google just stops citing you and cites another site in your place. For a younger site, that citation — even without a guaranteed click — is brand visibility you cannot get any other way today. And since the report has no clicks of its own, you have no way to prove with data that AI is actually costing you traffic; you would be making a permanent call based on a hunch.

The decision does make sense for a different profile: publishers and sites that monetize exclusive content directly, where every visit lost to a generated summary carries a measurable revenue cost. If that is you — or if you need to defend the decision to a skeptical client — here is the protocol we recommend, kept fully reversible throughout.

This experiment needs a full six to eight weeks on Pro: the 7-day Pro trial is not enough to cover even the baseline, and the Free plan's weekly check gives you just 4 readings a month — not enough resolution to pin down when each folder actually started moving. Everything measured continuously throughout the protocol — baseline, daily checks, folder comparison — is summarized in Rankiamos's daily rank tracking by city.