Google has finally given publishers a way to see what was previously invisible. On June 3, 2026, the company announced dedicated Search Generative AI performance reports in Search Console, shown in the interface as the Generative AI Features Report (Beta), with separate views for Search and Discover, showing how often a site’s pages appear inside generative AI features such as AI Overviews and AI Mode.
The reports first reached a subset of websites for testing and feedback, and as of August 31, 2026, Google says they have rolled out to all websites worldwide. For anyone whose traffic depends on search, this is the moment AI visibility stops being a guess and starts being a number.
What The Reports Actually Measure
The new reports track one thing above all else: impressions inside generative AI features. An impression here means a link to your site was shown to a user inside an AI Overview, an AI Mode answer, or a generative AI feature in Discover.
That is a narrower and more specific measure than the main Search Console performance report, which blends every kind of search appearance together. The dedicated view exists so publishers can separate their AI-driven visibility from their traditional blue-link visibility and watch each one move on its own.
The Search report covers two generative AI capabilities: AI Overviews, the AI-generated summaries that appear above or alongside classic results, and AI Mode, the conversational search experience where users ask follow-up questions and the context carries from one question to the next. Google says it expects to update that list over time as Search develops. Discover gets its own separate report, because Discover is a feed rather than a query-driven results page, and the way content surfaces there behaves differently.
One important detail is where the data lives. The generative AI data is included in the overall performance report, so a site’s total visibility picture stays intact, but the dedicated view is where the AI-only story can be read clearly.
If your site has not received enough impressions inside generative AI features, the report may not appear at all, which is itself a signal: it tells you that AI surfaces are not yet a meaningful source of exposure for your content.
From Experiment To Measurement
The road to these reports started years before they arrived. In May 2023, Google introduced its Search Generative Experience (SGE) as an experiment in Search Labs, offering AI-powered snapshots of key information with links to dig deeper and follow-up questions that opened a conversational mode.
That experiment grew into AI Overviews, and then into AI Mode, as generative AI moved from a side feature to a central part of how people search. For most of that journey, publishers could see their classic rankings in detail but had no dedicated way to see their AI presence.
The June 2026 announcement closed that gap, and the August 2026 worldwide rollout made the measurement universal. The timing matters because it changes the conversation between publishers and Google. Debates about AI answers and publisher traffic no longer have to rely on estimates or third-party studies. Every site owner with Search Console access can now open the report and see their own AI impressions, their own top pages, and their own country and device patterns.
Impressions: The Core Metric
Impressions are the foundation of the whole report. They answer a simple question: how often did links to your site get shown inside a generative AI feature? The chart at the top of the report shows total impressions for your property over the selected period, and the newest data can be preliminary, marked with a dotted line, because it is still being collected and may change over the next few hours.
How impressions are counted deserves attention, because the chart and the table do not always agree. Chart data is aggregated by property, so if two results from the same site appear inside one AI feature, they count as a single impression in the chart total. Add a URL filter and the chart aggregates by URL instead.
The table groups data by whichever dimension you select, and data grouped by country, device, or date is aggregated by property, while data grouped by page is aggregated by page. Those differences explain why chart totals and table totals sometimes differ, and Google points publishers to the same troubleshooting guidance used for the regular performance report.
Pages: Which Content Gets Cited
The Pages dimension is where strategy begins. It groups impressions by the final URL linked by a generative AI feature after any redirects, and like the main performance report, most data is assigned to the page’s canonical URL rather than a duplicate. Sorting this table shows which of your pages earn the most AI impressions and which earn the least, and that ranking is often very different from your classic search ranking.
A page that ranks fifth in classic results can be the most-cited page in AI answers, because AI features reward direct, complete, well-structured answers rather than the strongest backlink profile. Conversely, a page that dominates classic rankings may barely appear in AI features if its content is hard to extract, buried under clutter, or written for keywords rather than questions.
The Pages table turns those hunches into a list you can act on: strengthen the pages that already get cited, and diagnose the pages that should be cited but are not.
There is also a discipline hidden in this dimension. Because data follows the canonical URL, duplicate and near-duplicate pages split nothing and confuse nothing; the report shows the page Google actually treats as the source. Cleaning up canonical signals, redirects, and thin duplicates is therefore not just classic SEO hygiene, it directly determines whether your AI impressions are attributed to the right page.
Countries: Where AI Visibility Lives
The Countries dimension groups impressions by the country where the search originated. For publishers with international audiences, this is one of the most revealing views in the report. AI features do not roll out evenly across languages and regions, and a site can be heavily cited in one country while being nearly invisible in another, even for the same content.
Country patterns point to concrete decisions. A strong showing in one market may justify localized versions of your best-cited pages, while a weak showing elsewhere may reflect language coverage, regional competition, or simply the pace of AI feature availability in that market. Comparing the Countries table against your classic search country data shows whether your AI visibility mirrors your traditional reach or diverges from it, and divergence is where the opportunity usually sits.
Devices: How People Meet AI Answers
The Devices dimension groups impressions by desktop, tablet, or mobile. Google notes that device data is available for Search results, and the split matters because AI search behaves differently across screens. Mobile users lean on quick answers and voice-style queries, while desktop users run longer research sessions with follow-up questions in AI Mode.
If your AI impressions skew heavily mobile, your content needs to survive small screens and fast answers: short paragraphs, clear headings, and conclusions that hold up when only part of the page is read. If desktop dominates, longer and deeper pages may be doing the citing work, and the follow-up-question structure of AI Mode becomes more important.
Device data also interacts with page design, because a page that is slow or awkward on mobile may still get cited, but the click that follows the citation is less likely to happen.
Dates: Watching AI Visibility Over Time
The Dates dimension groups impressions by day, week, or month depending on the granularity you choose, and the report supports hourly, daily, weekly, and monthly views. All dates use Pacific Time, matching the main performance report, and the same rules about time zones and granularity apply.
Time-based views turn the report from a snapshot into a monitoring tool. A sudden rise in impressions after publishing a new guide, a drop after a site change, or a steady climb across a quarter each tell a different story, and each one can be checked against what you actually did on those dates.
Seasonal patterns show up here too: AI visibility for some topics spikes around events, product launches, or news cycles, and knowing that rhythm helps you publish before the spike rather than after it.
Search Types: Text And Multimodal
A filter at the top of the report lets you split impressions by search type. Text-based search covers queries typed into the standard search bar, while multimodal search covers queries made with images, including Google Lens, Circle to Search on Android, uploaded images, and Chrome image search. That distinction is easy to overlook and hard to overstate.
Multimodal AI impressions mean your images, product photos, diagrams, and visual content can be the entry point into an AI answer, not just your words. Sites with strong original visuals, clear image metadata, and pages where images genuinely explain something have a second route into AI features that text-only pages lack. If your multimodal share is zero, that is a finding in itself: your visual content is not yet participating in AI search.
Discover: A Separate Stage
Discover deserves its own paragraph because it is a different kind of visibility. Search answers a question the user asked; Discover suggests content the user might want before they ask anything. Generative AI features inside Discover therefore reward freshness, strong imagery, and broad appeal in ways that query-driven AI answers do not.
Google provides a dedicated Discover version of the generative AI performance report rather than folding it into the Search view, and publishers should read the two separately. A page can perform well in Discover’s AI features while barely registering in Search’s, or the reverse, and blending them would hide both stories. Treat them as two stages with different audiences and different rules.
Why It Matters: Search Is Shifting From Links To Answers
The deeper significance of these reports is not the dashboard, it is what the dashboard confirms. Generative search is shifting Search from a list of links toward answers delivered in conversation. Users increasingly get what they need inside an AI-generated answer, sometimes without clicking through to any website at all.
That changes what ranking means: appearing tenth in classic results used to be the goal, and now being featured inside the AI answer itself is a separate and often more valuable prize.
This shift punishes a specific kind of content. Generic, keyword-stuffed pages written to rank rather than to inform are exactly the pages AI systems have the least reason to cite, because they add nothing the model cannot already say. Original reporting, first-hand experience, clear data, and genuine expertise are what get quoted, because they give the answer something it did not already have.
The reports make that judgement visible: pages built on real value accumulate AI impressions, and pages built on optimization tricks do not.
The honest consequence is that impressions are not clicks. A page can be cited in thousands of AI answers while its click-through stays flat, because the user got the answer without visiting. Publishers who only track clicks will misread their own success, and publishers who only track impressions will overstate it. The two numbers have to be read together, and the new report finally supplies the second half of that picture.
How To Use The Reports Every Week
A practical weekly routine turns the report into decisions. Start with the Pages table and note your top ten cited pages, then ask what those pages have in common: their structure, their length, their use of headings, lists, and direct answers. Apply those patterns to the next pages you publish. Then check the bottom of the table for pages you expected to see cited but that are missing, and diagnose them individually rather than guessing at site-wide causes.
Next, compare countries and devices against your classic performance data. A country where AI impressions lag classic impressions is a localization or coverage opportunity. A device skew tells you where to invest in page experience. Then look at the Dates view around your publishing calendar and your site changes, so you learn which actions actually moved AI visibility and which ones only felt productive.
Finally, export the data. The report offers an export button for both chart and table data, and values shown as unavailable in the interface become zeros in the download. Building a simple spreadsheet from monthly exports gives you a history that survives interface changes and lets you compare quarters, campaigns, and content refreshes on equal terms. The usual data limits apply, including the 1,000-row cap, so export before a view gets too large to hold comfortably.
What The Reports Do Not Show
Understanding the limits prevents misreading the data. The generative AI performance report is built around impressions, and it does not replace the main performance report for clicks, click-through rates, or query-level analysis. A cited page’s journey from impression to visit still needs the classic report, analytics, and on-site measurement to complete.
The report also reflects only what Google’s AI features chose to show, not why. It will not tell you which sentence earned a citation, which competing page was preferred, or how the model weighed your content against others. Those answers still come from reading the AI answers themselves, comparing cited sources, and improving the underlying content. The report tells you where to look; it does not do the looking for you.
There is one more limitation worth stating plainly. Because the report can be absent for sites with too few AI impressions, a missing report is not a bug to troubleshoot first. It is usually a content signal: the site has not yet produced pages that AI features want to cite, and the fix is editorial rather than technical.
Optimizing Your Website For Generative AI Features
Generative engine optimization is the name this shift has acquired, and the reports give it an evidence base. The playbook that follows from the data is straightforward, even if executing it is not easy. Create content that answers real questions clearly, with the answer visible early and the supporting detail arranged underneath it.
Structure pages so machines and humans can both follow them: descriptive headings, short sections, lists where they fit, and questions answered directly rather than buried. Semantic understanding sits underneath all of this: AI features cite pages whose meaning is clear, not pages that merely repeat a keyword.
Authority and trust decide which pages get cited when several could answer. Demonstrated experience, named authors, original data, and a record of accuracy all feed the judgement that a page is worth quoting. That is the substance behind the familiar experience, expertise, authoritativeness, and trustworthiness standard, and AI citation makes it measurable rather than theoretical. A page that merely repeats what ten other pages say gives an AI system no reason to choose it.
There is a related discipline for AI Mode specifically. Because AI Mode carries context across follow-up questions, content that anticipates the second and third question earns more than content that answers only the first. Pages organized around a topic, with its natural follow-ups covered in their own sections, fit conversational search better than pages built around a single keyword.
Internal linking matters here too: a page that connects to the rest of your coverage on the topic helps both readers and systems treat your site as a coherent source rather than a scattered one.
As an AI optimization guide, this report rewards a simple loop: publish a clear answer, check whether AI features cite it, and improve the pages that do not get cited. That loop beats any checklist, because it uses Google’s own measurement rather than third-party guesses about what AI search wants.
Frequently Asked Questions
Where do I find the reports? Inside Search Console, under the performance section, look for the Generative AI Features Report (Beta) view, offered separately for Search and Discover. If it is not visible, the usual reason is that the site has not yet earned enough AI impressions for the report to populate.
Which generative AI features does the Beta report cover? The Search report covers AI Overviews and AI Mode, with Google saying the list will grow over time. Discover has its own version of the report for generative AI features inside the feed.
Do AI impressions count in my overall performance data? Yes, the generative AI data is included in the overall performance report, and the dedicated view simply separates it so AI visibility can be tracked on its own.
Why do chart and table totals sometimes differ? Because aggregation differs: chart data is aggregated by property, while table data grouped by page is aggregated by page, and data grouped by country, device, or date is aggregated by property. Small differences from aggregation are expected.
Can I export the data? Yes, both chart and table data can be exported, with unavailable values appearing as zeros in the download. The standard performance report limits, including the 1,000-row cap, also apply.
Is this the end of classic SEO? No, but it changes what classic SEO is for. Rankings still matter, and the pages that rank well often supply the material AI answers cite. What has changed is that ranking alone is no longer the whole game: being cited inside the answer is now a measurable outcome of its own, and the publishers who track both will make better decisions than those who track only one.
Background
The story of AI in Search runs from the 2023 Search Generative Experience experiment, through AI Overviews and AI Mode, to the June 2026 launch of dedicated measurement and its worldwide rollout on August 31, 2026. Each step moved generative AI further into the center of search, and each step made publisher visibility harder to see with old tools alone.
The new reports close that gap, and the publishers who learn to read them early will have the clearest view of where search is going next. For broader context on how answer engines are changing discovery, see how answer engines compare with classic SEO, and for the legal pressure shaping AI answers, see the German court ruling on AI Overviews and AI Mode.
The citation risks of AI search are covered in the study of AI tools citing fictional concepts, and the traffic stakes in the analysis of AI assistants and Google organic traffic. Our topics index gathers the full range of our coverage.
0 Comments