AI Mode Search Console reporting: what the generative AI report answers, and what it doesn't

AI Mode Search Console reporting exists, it is called the generative AI performance report, and it counts exactly one thing: impressions. There is no search term dimension, no clicks column, no average position. AI Overviews and AI Mode are folded into the same number. Almost every analysis people want from it has to be rebuilt at the page level, by hand, from the four dimensions it does give you.

Read this first

You need a verified Search Console property, and the report has to have reached it. Google's help page is blunt about that: "We're rolling out this report to a subset of website owners, allowing for thorough testing before rolling it further," and, further down, "Not all properties have access to the report, as we're rolling out over time." If it is not in your left-hand menu, skip to the last section, which is written for that case.

You also need something to compare against. A report that shows only impressions is close to useless without a before, which is why the handbook asks you to record your baseline before you change anything two steps earlier.

What AI Mode Search Console reporting actually contains

Four dimensions, one metric. That is the whole surface, and knowing the shape of it saves you from hunting for tabs that were never there.

Two exclusions are stated outright. Search Labs is out — "Search Console doesn't include data from experiments in Search Labs, as these experiments are still in active development." And the ordinary Search Console ceilings still apply: "the usual data limitations (1,000 row limitation, time period, etc) for the Search performance report also apply to this report." Every quotation in this chapter was read on 2026-08-19, from the generative AI performance report help page at support.google.com and the June 2026 launch post on the Search Central blog at developers.google.com.

A report with no search term dimension cannot tell you why you were shown. It can only tell you that you were.

Why the missing search terms change what you can conclude

Without search terms, you lose the ability to attribute a change to a topic. You can see that a page gained impressions in AI features last week. You cannot see that it gained them because people started asking a different question. Those are different findings, and only the second one tells you what to write next.

This matters more here than it does in ordinary search reporting, because AI answers are assembled from long, conversational phrasings that rarely match the wording on your page. In classic Search Console reporting you could at least read the wording people used. In this report you get a URL and a count.

What one flat count cannot separate

Here is what that costs in practice. Say a comparison page picks up 4,000 AI impressions in a month, against 900 the month before. Three explanations fit that shape equally well: the page started surfacing for a topic you never targeted, a competitor's page dropped out and you inherited its exposure, or AI features simply expanded into a category where you already ranked. The report distinguishes none of them, and the three call for three different next moves — write more of that topic, defend the position, or do nothing at all.

The dimension worth more than it looks

The countries dimension is the one that repays attention, and it gets overlooked because it looks like a vanity slice. If you sell into several markets from one site, it will show you that your AI exposure is concentrated in one or two of them long before your revenue numbers do. That is a real finding you can act on, and it is available today.

There is a second-order effect worth naming. Google's announcement says the generative AI data "is included in the overall performance report, where it will continue to be tracked to give site owners an overview of the overall visibility of their site in Google Search." So the AI numbers are inside your headline totals as well as broken out here — which means you cannot subtract one from the other to get a clean non-AI baseline. There is no view that shows you web results with AI stripped out.

How to get page-level answers anyway

The workaround is to stop asking the report for topics and start asking it for pages, then supply the topic yourself from what you already know each page targets.

1. Export the Pages table for a fixed window — 28 days is a reasonable default because it survives weekday effects. Use the export button; values shown as ~ or - come out as zeros in the download.
2. Join each URL to the primary search term you assigned it when you published. If you kept the one-line-per-page record from the cannibalization chapter, this is a lookup rather than a guess.
3. Group the impressions by that assigned term. You now have a topic-level table the report refused to give you, built on your own labels.
4. Repeat for the previous equal-length window and diff the two. The absolute numbers matter less than which rows moved.
5. Only then look at Countries and Devices, and only for rows that moved. Slicing a flat row by country produces noise.

Step 2 is the whole trick, and it is only cheap if you did the bookkeeping earlier. If you did not, the join is a manual afternoon for a hundred-page site and not worth attempting for a thousand.

The deliverable: which questions the report can answer

Work down this table before you open the report, not after. Three of these rows send you somewhere else entirely, and knowing that in advance is most of the value.

Three ways this report gets misread

Each of these produces a number that looks like an answer and is not one.

1. Reading the chart total as a sum of the table. They are aggregated differently. Google spells it out: "if two results from the same site appeared in a generative AI search results feature, they count as a single impression in the chart total." Chart is per property, page rows are per page, and the two will not reconcile.
2. Treating the newest days as final. The most recent data is preliminary, drawn with a dotted line on the graph, and it moves. Judging a launch on yesterday's figure is judging a number that has not finished arriving.
3. Adding the AI impressions to your total impressions. They are already inside it. Doing the addition double-counts, and it double-counts most on exactly the pages you care about most.

Where this chapter stops

We do not have this report on a property of our own, so everything above is read from Google's documentation and its announcement rather than from a screen we can see. If your rollout differs from what the help page describes, believe your own account.

We also cannot tell you how impressions in AI features convert to visits, because the report has no clicks column and Google has not published a conversion rate. Anyone quoting one is quoting an estimate from their own property. Google says it is "continuing to work with website owners to understand what insights and data would be most helpful," and mentions "adding additional metrics over time" — so this shape is current, not permanent.

Two more limits. This is Google only: it says nothing about being cited by other answer engines, and there is no equivalent report from any of them. And a property with thin traffic may simply see the empty state, which reads "Your site hasn't received enough impressions in generative AI features on Google Search" — that is a volume message, not a diagnosis.

If the report has not reached you yet

Do the page-level version by hand and it keeps working when the report arrives. Pick the ten pages you most want cited, record their current search impressions and average position from the ordinary performance report, and note the date. That is your before.

Then check the same ten pages against the fourteen-day method in how to tell if an SEO change worked. The mechanics do not change when a new report appears; you just gain one more column. Keeping that habit across a site, so the before is always there when you need it, is what QueryWin is being built to do.

One caution on eligibility. If your site is excluded from generative AI features, the report will stay empty no matter how the rollout goes, and the empty state will not distinguish "excluded" from "too small". Check your exclusion setting before you conclude anything from a flat line.

Common questions

Does the generative AI performance report show search terms?

No. The documented dimensions are pages, countries, dates and devices. There is no search term dimension, and Google has not said one is coming.

Does it show clicks from AI Overviews?

The report documents impressions. Clicks are not among the dimensions or metrics described on the help page, and Google frames additional metrics as something it may add later.

Can I separate AI Mode from AI Overviews?

Not in this report. Both are listed as capabilities included in the same impression count, with no dimension that splits them.

Why don't my AI impressions match my total impressions?

Because the generative AI data is also counted inside the overall performance report rather than sitting beside it. The two are not meant to add up.

How far back does the data go?

The help page points at the standard performance report rules for time period and row limits rather than giving a separate history for this report. Read your own date picker; it is the only reliable answer.