Google has finally given site owners a first-party view of where their pages appear in AI Overviews and AI Mode. That is useful. But the new Generative AI performance report is narrower than its name makes it sound, and reading too much into one number can send a team in the wrong direction.
The report measures qualifying link impressions. It does not currently isolate clicks, prompts, leads, sales, or the difference between an AI Overview and AI Mode. So the real job is not simply finding the report. It is knowing what the data proves, what it cannot prove, and which other numbers belong beside it.
Reporting note: This article reflects public Google documentation available on August 12, 2026. The report is still being released gradually, and its metrics, availability, and interface may change.
Key takeaways
- The dedicated report shows impressions from AI Overviews and AI Mode, with breakdowns for pages, countries, devices, and dates.
- It does not currently show prompts, clicks, click-through rate, average position, conversions, or separate totals for AI Overviews and AI Mode.
- The data already sits inside the regular Web search totals, so adding the dedicated report to overall impressions would count the same visibility twice.
- An impression means a qualifying link was shown under Google’s counting rules. It is not a visit, recommendation, citation rank, or unique person.
- The most useful review pairs this report with regular Search Console data, analytics, page changes, and a small, consistent prompt sample.
What changed in Google Search Console?
Google announced the dedicated Generative AI performance report for Search on June 3, 2026. The report pulls impressions from AI Overviews and AI Mode into one focused view. Before that, the same activity was folded into the Web search type in the normal Performance report, where it could not be inspected on its own.
As of August 12, 2026, Google’s help page still describes access as a rollout to a subset of website owners. That wording matters. If the report is missing from your property, it does not automatically mean the site has no AI visibility or that something is broken.
This is a better first-party baseline, not the prompt-level attribution report many teams hoped for. Google gives you a cleaner count of link appearances and enough dimensions to find page-level patterns. It stops well before explaining the full path from a generated answer to a customer.
What does the report actually measure?
The report currently has one metric: impressions. Google defines that as the number of times a link to your site was shown to a user in a supported generative AI feature. In an AI Overview, the link must be scrolled or expanded into view before it receives an impression. AI Mode follows Google’s standard impression rules.
You can break that impression count down four ways. Pages shows the final linked URL after redirects, usually credited to Google’s selected canonical URL. Countries shows where the search began. Devices separates desktop, tablet, and mobile. Dates can be viewed hourly, daily, weekly, or monthly, depending on the selected range.
| The report can answer | The report cannot answer |
|---|---|
| Which pages appeared | Which prompt or query triggered them |
| Whether impressions rose or fell | Whether AI Overviews or AI Mode caused the change |
| Which countries generated impressions | Whether the viewer was a qualified prospect |
| Which devices were used | What exact answer text surrounded the link |
| How visibility changed by date | Whether an impression led to a conversion |
And there is one accounting rule to keep straight: these impressions are already included in the regular Search Performance report under Web. Do not add the two totals together. The dedicated report is a filtered view of part of the whole, not a new pool of visibility sitting beside it.
What does an impression not mean?
An impression is useful, but it is easy to oversell. It does not mean someone visited the site. It does not mean Google recommended the company. It does not identify a unique person, prove that a passage was quoted, or show that a page held the most prominent supporting link.
- A click or visit
- A lead, purchase, or other conversion
- A unique person
- A recommendation from Google
- A citation rank or share-of-voice score
- Proof that a page edit caused later growth
Google does count qualifying clicks from AI Overviews and AI Mode inside ordinary Web search data. The problem is that the dedicated Generative AI report does not currently isolate those clicks. A page can record 2,000 generative Search impressions, and this report alone cannot tell you whether those appearances produced two visits or two hundred.
The same restraint applies after a content update. If impressions rise two weeks after a rewrite, the timing is worth recording. It is not proof that the rewrite caused the increase. Search demand, rollout changes, competitor pages, indexing, and Google’s own systems can all move at the same time.
Why can chart totals and page totals disagree?
Search Console can aggregate the same result differently depending on where you look. At the property level, one generated result containing two pages from the same site can count as one site impression. In the Pages table, each qualifying linked page can receive its own impression. That means the page rows may add up to more than the total in the chart.
Here is a hypothetical example. An AI response shows a maintenance guide and a pricing page from the same plumbing company. The property chart records one impression for the site, while the Pages table records one impression for each URL. The chart says one. The table rows add up to two. Neither is necessarily wrong.
A few other limits can create confusion. The table displays up to 1,000 rows. Recent data may be preliminary when Google marks it with a dotted line. Values shown as a tilde or a dash become zero in an export. And Search Labs experiments are excluded. A mismatch deserves a closer look, but it is not automatically a tracking defect.
Why don’t I see the report?
Google lists three common reasons. Your property may not have received access yet. The site may not have enough impressions in supported features. Or someone may have excluded the site through the Search generative AI control.
Start in Search Console Settings and check that control before assuming the worst. Google says inclusion is allowed by default. Then confirm that the pages you care about are indexed and eligible to show a snippet. If those basics are sound, a missing report may simply be a matter of access or volume.
But do not use the absence of this report as a site-health diagnosis. It is not a crawl report, indexing report, or manual-action notice. Search Console already has separate tools for those jobs.
Build a baseline before changing pages
A report this new creates a strong urge to start rewriting. Resist it. If you change ten pages before recording the starting point, you will have a pile of activity and no clean way to judge what followed.
Use 90 days when the property has enough history. If it does not, use the longest stable window available and state the range in every update. Then save the chart and table data so you have a record even if Google changes the interface later.
- Record total generative Search impressions for the period.
- Export the page, country, device, and date tables.
- Calculate how much of the total sits with the top five pages.
- Group visible pages by purpose, such as service, guide, category, product, or location.
- Mark major publishing, migration, and page-update dates.
- Compare the same pages in the regular Web search report.
- Review organic landing-page engagement and conversions in your analytics platform.
That last comparison is directional. The dedicated report and your analytics platform do not create a clean user-level join that proves an AI impression became a form submission. You are looking for patterns that deserve a test, not manufacturing certainty the data cannot support.
A hypothetical 90-day example
The following numbers are hypothetical. A B2B service company records 38,000 generative Search impressions over 90 days. One comparison guide receives 12,400, about 33 percent of the total. All service pages combined receive fewer than 2,000. Mobile accounts for 71 percent, and 84 percent of impressions come from the United States.
The sensible reading is modest. The guide is the company’s clearest current visibility asset in Google’s generative Search features. The team should check why it works, whether its facts are current, and whether an interested reader can reach a relevant service page without hunting. Stronger internal links and a clearer next step may help the page do more useful work after a visit.
What the team cannot say is that the guide produced 12,400 visits, that Google recommended the company 12,400 times, or that those impressions caused sales. It also would not be sensible to copy the guide’s format across fifty thin pages. One strong pattern is a reason to inspect, not a license to mass-produce.
Now picture a local HVAC company whose emergency-repair guide holds most of its impressions. Search Console still provides no prompt data in this report. The next move is to compare regular query data, read the page on a phone, and collect the questions real callers ask. Inventing a list of AI prompts from an impression total would only make the report look more precise than it is.
What should you do with strong and weak pages?
Start with the pages already appearing. Check dates, named sources, original examples, internal links, and the path to a relevant product or service. Make the central answer easy to find in visible text. If the page contains an old figure, an unsupported claim, or a dead end, fix that before adding more content.
For pages with little or no visibility, go back to ordinary search basics. Confirm crawling and indexing. Check whether the page is eligible for a snippet. Look at internal links, duplication, page experience, and whether the topic is useful enough to deserve a page at all.
Google’s published guidance says there is no special markup required for AI Overviews or AI Mode, and no special file needed for eligibility. Google Search does not use llms.txt as a ranking or visibility signal. That statement is about Google Search, not every other AI service. Clear pages, accurate structured data where it fits, useful media, sound technical access, and original information are still the work.
Those practices support eligibility and make a page easier for Google Search to process. They do not guarantee an AI Overview, an AI Mode appearance, a ranking, or a lead.
A monthly review can take 30 minutes
You do not need a giant dashboard. A short monthly check is enough for most teams, especially while access and reporting are still changing.
- Five minutes: compare the latest month with the prior month and, when useful, the same period last year.
- Ten minutes: inspect the pages with the largest absolute changes.
- Five minutes: check country and device shifts.
- Five minutes: review annotations for site and content changes.
- Five minutes: write one action, one open question, and one item that needs more data.
Every update should answer a short set of questions: What changed? Which pages moved? What might explain it? What can we verify? What will we watch next? If the report cannot support an answer, say so. A blank cell is better than a confident guess.
How does this change GEO measurement?
Sampled prompt tracking still has a role, but it should not carry the whole measurement plan for Google’s generative Search features. Search Console now supplies first-party impression history. A controlled prompt sample can add question and competitor context. Analytics can show what visitors do after they reach the site. None replaces the others.
The same separation helps when teams measure social and video content in Search Console. Platform properties show how supported off-site accounts perform in Google Search, while the Generative AI report covers qualifying links from a website inside supported AI features. They answer related questions, but they are not interchangeable.
Our view is simple: use the Generative AI report as a better flashlight, not a scoreboard. It can point to pages and shifts worth examining. A sound measurement plan still needs clear definitions, consistent dates, normal search data, analytics, and someone willing to write down what the numbers do not show.
If the report is collecting data but your team still cannot explain which pages matter or what should happen next, Ardoz Digital can review the page patterns, compare them with wider search and analytics data, and identify practical next steps. That work can produce a clearer plan. It cannot promise placement in AI Overviews or AI Mode.
Ardoz Digital is not affiliated with or endorsed by Google. Google product names are used only to identify the products discussed.