Ana Fernández / SEO

Search Console AI Report: What It Measures and How to Use It

Google rolled out Search Console's generative AI performance report to every site. What it measures, what it leaves out, and how to read your AI impressions.

September 4, 2026 7 min readby Ana Fernández

For almost two years, the most repeated question in my meetings with marketing teams was some variant of "how do I know if I show up in Google's AI answers?". The honest answer was frustrating: there was no direct way to know. Google mixed that data into the general Search Console report without separating it, and what was left was estimating with third-party tools or running searches by hand like it was 2005.

That just changed. On August 31, Google finished rolling out the generative AI performance report in Search Console to every site in the world. For the first time you can see how many times your site appeared in AI Overviews and AI Mode, with official data, for free. And along with the report came a new button that deserves its own conversation: the option to exclude your content from Google's AI answers.

Let's go step by step, because the report is useful, but it comes with fine print.

What is the generative AI performance report?

It's a new report inside the performance section of Search Console, separate from the classic search results report. Google announced it in June 2026 and released it gradually over the following months, until confirming in late August that it's now available globally.

The report shows your performance in two surfaces: AI Overviews (the AI summaries that appear above the results) and AI Mode (the tab where search works as a conversation with a chatbot). There's a separate report for AI features in Discover, and Search Labs experiments are left out of the count.

One practical detail before you rush to look for it: even though the rollout is officially global, Google's help pages acknowledge it's still in progress, and if your site has few appearances in AI features, the report may not show up for you yet. If you don't see it, check again in a few days.

What exactly does it measure (and what doesn't it)?

Here comes the fine print. The report measures a single thing: impressions. That is, how many times a link to your site was shown inside a generative AI feature. You can segment those impressions by page, country, device and date, with the same logic as the regular performance report, including the 1,000-row limit.

What it does NOT include is just as important. There are no clicks, no CTR and no position. In other words, you know you appeared, but you don't know whether anyone reached your site thanks to that appearance, or where in the answer you were.

Why does that absence matter so much? Because user behavior in an AI answer differs from the classic results page. In an AI Overview, a good share of people read the summary and leave without clicking anything. Having impressions without clicks tells you your brand is present at the moment someone asks, and that presence is worth something, but you can't convert it directly into traffic or a CTR curve.

My reading is that this report looks more like a share of voice metric than a performance metric. It helps answer "am I in the conversation or not?", and that connects with what I've written about the metrics CMOs should be looking at to measure organic: the part of the picture that organic sessions no longer capture.

How to start using it this week

The full recipe, no paid tools required.

1. Write down your baseline

Open the report (it lives under Performance, next to the search results one) and write down three numbers: total AI impressions for the last month, how many distinct pages appear, and what share of those impressions comes from AI Overviews versus AI Mode. That's your starting point, and without it any analysis you run six months from now will be pure guesswork.

2. Identify which pages are getting you into the answers

Sort by impressions and look at the top 10 pages. I bet the list won't match your highest-traffic pages. AI systems cite content that answers specific questions clearly, and sometimes that modest page nobody was watching turns out to be your best ambassador in front of the AI.

3. Cross-check with the classic report

Take those same pages and look at them in the traditional search results report. If a page has lots of AI impressions and its organic clicks have been dropping, you now have a reasonable hypothesis that the AI answer is absorbing part of those clicks. The report doesn't connect the two, so this stays at the hypothesis level, but it gives you a much better starting point than blaming the drop on chance.

4. Segment by country if you sell in several markets

The rollout of AI Overviews and AI Mode advances at different speeds depending on country and language. A client operating in Chile, Peru and Mexico can have completely different AI impression profiles in each market, and that data changes where it makes sense to invest in content first.

5. Set a monthly ritual

Once a month, compare against your baseline: AI impressions, new pages that entered the answers, pages that dropped out. Fifteen minutes. With six months of data you'll have your own time series that no external tool can give you, because this comes straight from Google.

What about the button to opt out of AI answers?

Along with the report, Google activated a property-level control in Search Console settings to exclude your content from AI Overviews, AI Mode and Discover's AI features. Two important official clarifications: using it doesn't affect your ranking in traditional results, and it doesn't stop Google from training models with your content either, because that's handled by a separate control called Google-Extended.

I understand the temptation, especially for publishers and content sites watching the AI summary answer with their material without sending them the visit. But before pressing the button, it's worth looking at the numbers in the new report and doing the math: how many impressions would you be sacrificing, and what role do they play in people getting to know your brand?

There's one piece of data that makes me very cautious about opting out. AI systems strongly favor brands they already know: a study from July measured that models search for familiar brands 3.2 times more often when building their answers, something I covered in detail in this article about AI's favorite brands. Leaving AI answers today means giving up exactly the space where that familiarity is being built. For most companies, which need to become more known rather than less, switching that off looks like a bad trade to me.

What this report doesn't solve

It's worth saying plainly: this measures your visibility in Google's AI features, and only there. ChatGPT, Perplexity, Claude and Gemini as an app remain outside any official report, and for those fronts you still need other methods: systematic prompt checks, AI crawler logs and referral traffic from the assistants.

The complete picture of your AI visibility comes from combining both sources. The good news is that the Google piece, which was the biggest and the most opaque, now has official data.

Wrapping up

If you manage a site, this week's homework fits in half an hour: open the report, write down your AI impressions baseline, identify your 10 most cited pages and schedule the monthly review. When someone at your company asks in three months "how are we doing on AI visibility?", you'll be the only person in the room with an official data series to answer with, and that already puts you several steps ahead of most of the market.

Free weekly resource

Get the enterprise SEO playbooks before anyone else

Every week we publish a new free resource: interactive audits, calculators, checklists and templates we actually use with enterprise clients. Leave your email and we'll send it to you first.

No spam. Unsubscribe whenever you want.

Enterprise SEO consulting

Want to apply this to your brand?

We design end-to-end SEO and GEO programs connected to pipeline and revenue. Senior team, weekly execution, executive reporting.