Polski
Content & AI search · News

AI reports in Search Console: what they show and what is missing

Google has added generative AI performance reports to Search Console: impressions in AI Overviews and AI Mode, broken down by page, country and device. No clicks, no CTR, no queries. I show how to draw sensible conclusions from them anyway.

Sebastian Baryś
Sebastian Baryś
1 August 2026 12 min read

Introduction

English version of an article first published in Polish on 1 August 2026. Read the Polish original.

For the past two years the question "how much traffic are AI answers taking from me" had to be settled by feel: clicks were falling while impressions held steady, so it was probably AI Overviews. In June 2026 Google finally provided a tool: generative AI performance reports in Search Console. In July it widened access and added documentation for the controls.

That is good news, with an asterisk half the size of the screen. The report shows impressions only. There are no clicks, no click-through rate and no queries. In other words, the very thing that would answer the question that started it all is missing.

I show what exactly you get, where the limits of this data lie and how to build a sensible picture from it all the same, including what to fill the gaps with.

You will learn:

  • Which surfaces the report covers and what is deliberately left out
  • Which dimensions you get (pages, countries, devices, dates) and what their limits are
  • Why the lack of an API and of a BigQuery export turns the 1,000-row limit into a real wall
  • Why the absence of clicks, CTR and queries is a serious interpretation problem
  • What to do when the report does not appear in your property
  • How Google’s data differs from what Bing has offered since February
  • A concrete way of working with data that shows impressions only

What the generative AI reports are

Google announced them on 3 June 2026 as separate views in Search Console: one for Search, the other for Discover. The aim is to show how your site performs in generative features, which until now were lumped in with the rest of the statistics and could not be isolated.

The Search report covers two surfaces: AI Overviews and AI Mode. Google adds that the list may grow as the product develops. Deliberately excluded are data from Search Labs experiments, generative features in Discover (they have their own report) and sites that have opted out of AI features.

One thing is essential for reading the numbers correctly: this data comes from the web search data in the performance report. The new report is therefore a slice of what you already have, not a separate pool of traffic. Adding one to the other is the easiest way to count the same impressions twice and hand a client a report that does not add up.

This is not new traffic that has suddenly appeared. It is a breakdown of traffic you already had, except that you can finally see what part of it passes through the AI layer.

What exactly the report shows

There is one metric: impressions, understood as the occasions on which links to your site were shown to a user in a generative feature. On top of that come four dimensions you can slice the data by.

The contents of the report in one place

What you get

  • impressions in AI Overviews and AI Mode
  • breakdown by page (canonical URL)
  • breakdown by country and device
  • dates: daily, weekly, monthly

What you do not get

  • clicks
  • click-through rate (CTR)
  • average position
  • query data
Four dimensions to slice the data by and one metric. The rest is still to come.

Then there are the limits familiar from the classic performance report: a 1,000-row limit in the table, preliminary data for the most recent days marked with a dashed line (it can change within hours) and a discrepancy between the chart total and the table total, which comes from different aggregation at property level and at page level. Dates are given in Pacific Time, which can shift the picture when you analyse day by day.

What is missing from it

The absence of clicks is not a detail to tick off in a footnote. It is the difference between "I know I was shown" and "I know whether it paid off". Lily Ray of Amsive summed it up briefly: reporting impressions instead of clicks "says a whole lot here". It is hard to disagree, since this very metric is at the heart of the dispute over AI Overviews.

The absence of query data hurts from the other side: you can see that a given page appears in AI answers, but you do not know for which questions. And that is exactly the knowledge that would let you write for citations deliberately instead of guessing. Without it, you are left with inference from the classic performance report and your knowledge of your own content.

AI impressions tell you that Google showed you. They do not tell you whether anyone clicked, whether they asked about something else entirely, or whether anything came of the visit.

There is no API or BigQuery export either

This limit is less obvious, and in day-to-day work it is the most annoying one. In the classic performance report the 1,000-row limit is purely an interface restriction, and for years there have been three ways around it: the Search Console API (up to 25,000 rows per request plus pagination through the startRow parameter), the Looker Studio connector or the bulk data export to BigQuery.

The problem is that the generative AI report has none of these routes. The type parameter of the searchanalytics.query method still accepts the same values as before (web, image, video, news, discover, googleNews) and no category for generative features has been added to them. There is no bulk export of this data to BigQuery either.

The practical consequence

The report lives only in the interface. You cannot build an automated dashboard from it, you cannot drop it into a recurring client report, and you cannot join it with conversion data in a warehouse. What remains is a manual CSV export, clicked through every month. On a large site the 1,000-row limit in the table therefore becomes a real wall, because here there is nothing to get around it with.

Google has said it will expand the set of metrics based on feedback from site owners. Until then, one rule is worth pinning above your desk: this is visibility data, not results data. Reporting growth in AI impressions to a client as a campaign success is about as honest as boasting about the number of banner impressions.

I cannot see the report in my property

The most common question since June. Three reasons, in this order:

  1. 01 The rollout. Google is making the report available gradually: the launch covered sites in the United Kingdom, and further markets followed. There is no switch that would turn it on earlier. All you can do is wait and check once a week.
  2. 02 Too few impressions. The report needs a sufficient number of impressions in generative features to have anything to show. Google has not published the threshold, but small local sites may simply not reach it.
  3. 03 Opting out of AI features. Rare, but worth checking: if the site has been excluded from generative features, the report will have no data source.

What a missing report does not mean: a penalty, a configuration error or dropping out of the results. The data in the report goes back to roughly 18 May 2026, so even once you get access the history will be short. All the more reason to export from day one.

A side note that shows how live this report is: on 22 July 2026 AI Overviews and AI Mode launched in France, and the French impression curves start exactly on that date. In international analysis it is easy to mistake a jump like that for your own success.

Google versus Bing: who gives you more

The comparison does not flatter Google and there is no point dressing it up. Bing started measuring AI visibility in February 2026, and in June it added metrics that Google still does not have.

What each tool gives you

Search Console

since June 2026

  • impressions in AI Overviews and AI Mode
  • pages, countries, devices, dates
  • a separate report for Discover
  • the scale of the market you actually operate in

Bing Webmaster Tools

since February 2026

  • citations in AI answers
  • Citation Share: your share among sources
  • Intents and Topics: why and what people ask about
  • a much smaller market share
Google has the scale, Bing has the metrics. That is why it makes sense to look at both at once.

The practical conclusion is simple: Bing shows the quality of your presence in AI answers, Google shows its scale in the market that actually sends you traffic. I have broken down the Bing metrics in detail in a separate piece on AI Performance in Bing Webmaster Tools. The account is free and takes a few minutes to set up, so it would be a waste to pass it up.

How to get value out of it

Impression-only data can still tell you a good deal if you set it against something else. Five things I do:

  1. 01 Calculate the AI share for each page. Impressions in the AI report divided by the impressions of the same page in the classic performance report, for the same date range. You get a rough indicator of what part of the page's visibility passes through the AI layer.
  2. 02 Look for divergence. Pages where the AI share is rising while clicks in the performance report are flat or falling are the best candidates for a zero-click effect. These are the pages that need a decision: rewrite for citation or rebuild for a different intent.
  3. 03 Slice by device and country. AI features behave differently on mobile and desktop, and they launch in different markets at different times. Without this split it is easy to mistake the launch of a feature in a given country for the effect of your own work.
  4. 04 Add sources from outside Google. Citation Share from Bing, referrals from assistants in your analytics, and server logs with AI bot visits. Only these three layers together give a picture that no single tool can give today.
  5. 05 Export a CSV every month, by hand. There will be no automation, because the data is neither in the API nor in BigQuery. The history is short, preliminary data changes and the row limit will cut the table. Your own archive is the only way to have a year-on-year comparison a year from now.
A single metric rarely means anything. The same metric set against a second one, from a different report, starts to tell a story.

AI controls and the nosnippet trap

Together with wider access to the reports, Google published documentation for the controls that let you limit the presence of your content in AI features. In practice these are directives that have been around for years: nosnippet, data-nosnippet, max-snippet and noindex. Their reach now extends to AI Overviews and AI Mode as well.

Before you reach for nosnippet

The same directive that excludes content from AI answers also removes your snippet from regular search results. Only the title and the URL remain, which can cut click-through harder than all the AI Overviews put together. Preview controls are a surgical tool: they work sensibly when applied selectively, to specific passages through data-nosnippet, not wholesale across the entire site.

It is worth remembering that outside the United Kingdom, Poland included, there is still nothing like what the regulator forced on Google there: a separate switch for AI features without losing regular results. I described this at greater length when covering the decision of the UK's CMA. Until that separation exists here too, the choice remains an uncomfortable one.

FAQ: common questions

What exactly does the generative AI performance report show?

It shows impressions, meaning the occasions on which links to your site were shown to a user in a generative AI feature in Search: in AI Overviews and in AI Mode. You can view the data by page (grouped by canonical URL, after redirects), country, device and date, at daily, weekly or monthly granularity. Google notes that the list of covered features may grow as Search develops.

Does the report include clicks and queries?

No. In the current version only impressions are available. Clicks, click-through rate, average position and query data are all missing. Google has said it will expand the set of metrics based on feedback from site owners, but for now you see what Search showed, not what brought a visit. So treat the report as an indication of direction, not as a measure of business results.

Are AI impressions counted in the regular performance report?

Yes, and this is an important thing to understand. The data comes from the web search data in the performance report, so the new report is a slice of what you already have in the standard statistics, not a separate pool of traffic. You should not add one to the other, because you would count the same impressions twice.

Why can I not see the report in my property?

The most common reason is the rollout: Google is making the report available gradually and some properties are still waiting. The second reason is too few impressions in generative features for the report to have anything to show. There is no switch that turns it on. A missing report does not mean a penalty or a configuration error, and opting out of AI features is an entirely separate setting.

How far back does the data go?

The history is short: the data starts at roughly 18 May 2026. The usual limits of the performance report also apply, including the 1,000-row limit in the table, and the most recent days are marked as preliminary and can still change within hours. Chart totals and table totals can differ, because aggregation at property level and at page level is calculated differently. The practical conclusion: export the data regularly instead of counting on it staying in the interface.

Can the AI report data be pulled through the API or exported to BigQuery?

Not at present. In the classic performance report the 1,000-row limit is only an interface restriction, and you get around it with the Search Console API (up to 25,000 rows per request, with pagination through the startRow parameter), the Looker Studio connector or the bulk data export to BigQuery. The generative AI report has none of these routes: the type parameter of the searchanalytics.query method still accepts only the values web, image, video, news, discover and googleNews, and there is no bulk export of this data to BigQuery. What remains is a manual CSV export from the interface, so on a large site the row limit becomes a real problem.

Is it worth adding Bing Webmaster Tools to this?

Yes, because Bing simply gives you more today. Since February 2026 it has measured citations in AI answers, and since June it has added Citation Share (your share of citations among all sources for a given query), Intents and Topics. Google shows that you were displayed; Bing shows what share you hold among the cited sources. The two sources complement each other, and a Bing Webmaster Tools account is free and takes a few minutes to set up.

Worth remembering

  • The report covers AI Overviews and AI Mode; data from Search Labs experiments is excluded, and Discover has a separate report.
  • The only metric is impressions: no clicks, no CTR, no average position and no query data.
  • This is not new, additional traffic: the data comes from the web search data in the performance report, so it is a subset of what you already have.
  • The report lives only in the interface: it is not in the Search Console API or in the BigQuery export, so the 1,000-row limit cannot be worked around the way it can in the classic performance report.
  • The history is short (data from roughly 18 May 2026), so export a CSV every month, because nobody will do it for you automatically.
  • A missing report in your property usually means the rollout has not reached you yet or there are too few impressions, not a penalty or a configuration error.
  • Bing gives you more today: Citation Share, Intents and Topics. It is worth having both sources and reading them together.

Summary

The generative AI reports are a step in the right direction and the first moment at which Google admits in its data that the AI layer is a separate reality. But in their current version they answer the question "was I shown", not "what do I get out of it". Those are two different conversations, especially when you are having them with a client.

We got an impression counter where everyone was asking for a click counter. It is better than nothing, as long as we do not confuse one with the other.

In practice: check the report regularly, calculate the AI share instead of looking at absolute numbers, export the data from day one, and add Bing and your own analytics to it. And when Google adds clicks, you will have a history to compare them with.

Want to know how much traffic the AI layer is taking from you?

I will put together the data from Search Console, Bing and your analytics, calculate the real AI share of your visibility and show which pages need rewriting. No reading of tea leaves.

Book an SEO consultation

How this text was written

I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.

What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.

Where the model helped

  • digging through sources and organising data
  • proofreading and the English version of the text
  • code, page layout and infographics

What stays with me

  • the choice of topic and what is worth saying about it
  • the arguments, the reading of the data and the conclusions
  • fact-checking and the decision on what goes into the text
Sebastian Baryś – freelance SEO specialist

Written by

Sebastian Baryś

Freelance SEO specialist · Częstochowa, Poland, and remote

I started when nobody had dreamt of Core Web Vitals and rankings were checked by hand. Today I work on making companies visible not only in Google, but wherever people ask questions. I write only about things I check myself in client projects.

SEOGoogle AdsTechnical SEOSEO in the AI era

Let's measure how much the AI layer really takes from you

Data from Search Console, Bing and analytics in one picture, with conclusions you can turn into decisions.

Get in touch