Search stopped returning a plain list of links, and measurement took a while to catch up. For more than a year, anyone watching organic traffic move had no way to tell whether the cause was ranking, seasonality or an AI answer served before the first click.
The AI report in Google Search Console exists to close part of that gap. It separates what used to arrive bundled together and shows how often a site's pages appear inside answers generated by artificial intelligence.
Through 2026 the panel moved from a narrow test to a broad rollout, and along the way it collected a reputation it does not deserve. The report answers fewer questions than it appears to.
This article walks through what the panel shows, what it stays silent about, and how to use each data cut without turning an impression into a hasty conclusion.
How does the generative AI report work in search?
The generative AI report consolidates the times a site's pages were displayed inside artificial intelligence features in search. It covers the AI-built Overview, the Google AI Mode and assisted answers in Discover, gathered into a dedicated panel.
Google introduced the report in June 2026, as part of a release that also brought new controls for site owners. The panel carries a Beta label and lives inside the Performance section.
The underlying change is separation, not volume. Before this panel existed, appearances inside AI answers already counted toward the general Performance report, blended with traditional organic data.
That means the number on screen is not new traffic. It is a slice that was always part of the total and only now gets its own view.
The history starts on May 18, 2026, and Google backfills the period that preceded each property's access. Properties added in later waves do not lose data: the series is already populated the first time the panel opens.
Caption: The AI report on GSC separates presence inside generated answers from the rest of search, showing impressions by page, country, device and day.
Which properties can already see the AI report on GSC?
Access to the AI report on GSC rolled out in waves through mid-2026 and became universal on August 31, 2026, when Google confirmed in its documentation that the report had reached every site.
The path was gradual. The panel started with a restricted group of creators in the UK for regulatory reasons, and by June 23, 2026 Search Engine Land reported access in the United States, India and Switzerland.
The widest wave arrived in August. The number of eligible properties grew sharply at once, and part of the industry read the change as a general release.
Not yet. John Mueller, from Google, addressed the confusion the same day: "Correct, it's not yet every domain. Also, like the discover and news reports, these are only shown if there's sufficient data to display."
The general release came at the end of that month. On August 31, 2026, the official Search Console documentation replaced the partial rollout language with an unambiguous confirmation:
"As of August 31, 2026, we've rolled out these insights to all websites worldwide."
The same date applies to the control that removes a site's URLs from AI-generated answers, released worldwide alongside the report.
Two practical consequences follow. There is no activation button and no queue: if the property is verified in Search Console, the panel is there.
The second is that an empty report stopped meaning "still queued" and now means "not enough volume". The panel only renders data once impressions cross the display threshold, the same way Discover and News reports behave.
Which data does the generative AI report show today?
The generative AI report shows impressions and nothing beyond that, spread across four cuts: pages, countries, devices and days. An impression is recorded when a URL appears as a source or grounding link inside a block of AI-generated content.
Each cut answers a different question, and it helps to know which one you are asking before opening the tab. Here is how it breaks down:
| Data cut | Question it answers | Most direct use |
|---|---|---|
| Pages | Which content AI uses as a source | Prioritize what already has traction |
| Countries | Where the presence concentrates | Check whether the right market is reached |
| Devices | On which screen the answer is read | Adjust content format and length |
| Days | When the presence shifted | Cross-reference with publishing dates |
Table: The four cuts available in the panel and the reading each one supports.
None of these cuts intersects with clicks, and that limitation defines how the report can be used. It describes presence, not outcome.
A scale example helps set expectations. On the mkt4edu property, on August 17, 2026, the report recorded 14.5 thousand impressions in generative AI features across a three-month window.
That figure does not represent visits, readers or leads. It represents how many times pages from the site were displayed inside an answer built by artificial intelligence.
Why does the report have no clicks or CTR for AI?
The report has no clicks or CTR because Google chose to publish only the impression layer at this stage. That gap was not closed by the June or August 2026 expansions, and no timeline has been announced for those metrics.
There is a distinction that confuses entire teams, and it deserves to be spelled out. Queries and clicks originating in Google AI Mode do exist inside Search Console.
They appear in the general Performance report, added to traditional organic data. What does not exist is the breakdown that would isolate them.
In practice, you know the total includes AI-driven search, but you cannot say how much of it came from there. That is the difference between missing data and unseparated data, and it changes what can be claimed in a results meeting.
While that layer is unavailable, external research helps size the effect. A Pew Research Center study published in July 2025 observed clicks on a traditional result in 8% of visits with an AI summary, against 15% of visits without one.
The same study recorded clicks inside the summary itself in 1% of visits. The sample covered 900 US adults with browsing tracked in March 2025, which calls for caution before generalizing the figure to other markets.
How do the GSC blocking control and Google-Extended differ?
The GSC blocking control is a toggle that removes a site's URLs from AI-generated answers. Once the opt-out is active, the domain stops grounding responses and stops appearing as a recommended link in AI Overviews, Google AI Mode and Discover.
Google clarifies a technical point that usually drives hesitation before the decision. Opting out of generative AI features does not affect how pages rank in traditional organic results.
The control coexists with an older mechanism, and the two solve different problems. Treating them as interchangeable leads to blocks that are broader than intended.
The Google-Extended agent, declared in robots.txt, prevents site content from being used to train language models such as Gemini. It acts on training, not on real-time display.
The panel toggle acts on the shop window instead. It decides whether the site appears as a source inside the answer a user is reading at that moment.
Choosing between them depends on what actually bothers you. Protecting content from training belongs in robots.txt, while leaving the answer belongs in the panel, at the cost of the citation and the brand exposure that come with it.
How do you diagnose organic traffic loss with this report?
The report helps diagnose organic traffic loss only when cross-referenced with the Performance report. The meaningful signal appears when a URL loses organic clicks while holding or increasing impressions in AI features over the same period.
That overlap suggests the answer is being delivered before the click. It does not prove causation, because ranking, seasonality and shifting search intent still compete as explanations.
Three misreadings show up repeatedly among teams opening the panel for the first time, and they are worth naming before any internal report is built:
- Adding AI impressions to the organic total. The number was already inside the general report, so adding it creates double counting.
- Comparing AI impressions with blue link impressions. The two come from different display events and are not equivalent, so a direct comparison supports no conclusion.
- Reading impression growth as audience growth. Appearing more often inside answers can coexist with fewer visits, and frequently does.
Understanding how Google's AI is affecting organic traffic helps calibrate expectations across the team and with leadership, before the conversation turns into pressure for a number the panel does not produce.
Until click data arrives, the most useful cross-reference comes from the CRM. Assisted conversions, direct brand traffic and growth in branded search tend to move when presence inside AI answers increases.
How do you structure content to appear in AI search?
There is no exclusive technical requirement to appear in AI search. Google states in its AI features documentation that there are no additional requirements to appear in AI Overviews or AI Mode: the page needs to be indexed and eligible for snippets.
That dismantles a good share of the promises circulating in the market. No magic markup, special file or secret format guarantees a citation, and working on Generative Engine Optimization means doing SEO with more rigor, not running a parallel discipline.
What is known about extraction comes from the mechanism, not from an official manual. AI search systems retrieve passages rather than whole pages, which favors text that stands on its own.
Three practices show measurable effect in studies of AI citation, and all of them depend on content rather than configuration:
- Cite named sources. Attribute the claim to an identifiable institution, with a year and a link, instead of leaving it floating in the text.
- Use verifiable statistics. A number with a declared origin survives extraction better than an adjective does.
- Include expert quotes. A direct, attributed statement gives a passage an anchor that paraphrase cannot provide.
One item aged badly and deserves a warning: the llms.txt file circulated widely as a recommendation and did not hold up. Google confirmed it does not use the file, and no correlation between adopting it and being cited has been demonstrated.
The work still lives elsewhere, closer to Search Everywhere Optimization: a crawlable page, clean HTML, the answer placed at the top of each section, and information that is not already available prepackaged somewhere else.
Frequently asked questions about the AI report in Google Search Console
So what does the AI report on GSC actually solve?
The AI report on GSC settles one question and postpones another. It answers where content appears inside generated responses, broken down by page, country, device and day, and leaves open how much of that becomes a visit, a lead or a customer.
Treating that boundary honestly is what separates productive use from an internal report that promises too much. An impression is presence, and presence is a condition, not a result.
Even so, a broader rollout changes the starting point of every conversation about traffic decline. Speculation about what artificial intelligence might be doing to organic performance gives way to primary data from Google itself.
The next step is to expand what the report merely measures. Appearing in more answers depends on topical authority, coverage of real questions and content that still holds value after AI has delivered the basic definition.
You know where you appear. And where you don't?
The report gives you the impression. Visibility is still built, and the path runs through how to increase visibility on Google.
.jpg)
Portugal