Search Console's generative AI report: how to read your AI Overviews and AI Mode impressions

You open Search Console, find the generative AI report, and there it is: one impressions chart, four tabs and not a single click. That is not a broken panel. That is everything Google chose to publish. Here is how to read it.

First what it measures and how it is counted, then its limits, and how to use it without inventing numbers.

What the generative AI report in Search Console actually is

It is a dedicated view inside your site's performance data showing the impressions your URLs earned in Google Search's AI features. Not sessions, not conversions: impressions.

The dates are verifiable: it was announced on June 3, 2026 with a rollout limited to a subset of sites, and since August 31, 2026 it is available to every site worldwide. Today it covers AI Overviews and AI Mode, with a list Google says it will expand. And there are two separate reports: one for Search and one for Discover.

The clarification that prevents half the confusion: the data is not new. It is a separate view of something already counted in the overall performance report, and the official documentation puts it plainly: those figures "remain included in the overall Performance report". This is not a traffic report. It is an exposure report.

What it measures: impressions in AI Overviews and AI Mode

An impression is how many times links to your site were shown inside an AI feature. It is not a visit, not a click, not a mention: it means your link was there.

What the report leaves out

There are no clicks, no CTR, no queries, no average position, and no text of the answer you appeared in. Search Labs experiments are excluded too, per the help documentation. The practical consequence is blunt: you cannot compute an AI CTR; any such figure is a division between two things that do not correspond. Google mentioned "adding additional metrics over time". Today they are not here.

Where to find the generative AI report and which tabs to use

It sits next to the performance report, with the same date filter and comparison controls. In Search you get four dimensions: Pages, Countries, Dates (day, week or month, always in Pacific Time) and Devices. Discover has no devices. Pages uses the final URL after redirects and assigns data to the canonical.

The search type filter: text or multimodal

Two options: Web: text, for typed queries, and Web: multimodal, which groups everything arriving as an image: Lens, Circle to Search on Android, uploaded images and Chrome image search. It separates exposure born from a sentence from exposure born from a photo. If you sell visual products, the second one matters; the mechanics are covered in multimodal search in Search Console.

How impressions are counted (and why the chart will not match the table)

Property aggregation vs page aggregation

The chart aggregates by property: if two results from your site appear in the same AI answer, they count as one impression. The table aggregates by whichever dimension you picked: on Pages, each URL counts separately. That is why the totals can differ with no loading bug. Do not add up the pages column to reconcile anything: chart for the trend, table to compare pages. It is what the official notes on impression counting describe.

Preliminary data and the 1,000-row ceiling

Recent data is preliminary: dotted line, and it can change in the following hours (today and sometimes yesterday). The table caps at 1,000 rows, so large sites have to export and filter outside. Watch the CSV: values you see as ~ or - become zeros, and a downloaded zero is not always zero impressions. For serious analysis, export to BigQuery.

The four limits that actually matter

1. No clicks, no queries. The question "which searches bring me AI Overviews?" has no answer here, nor in the Search Analytics API.
2. No position. You do not know where in the answer you appear, nor whether your link is visible or collapsed.
3. It is not exact accounting. Preliminary data, double aggregation and a minimum impression threshold make the figure a trend, not a balance sheet.
4. It depends on the generative AI control. If you excluded the site through that setting, you are no longer eligible and the report stays empty. Not a failure, the logical consequence.

With those pieces, the only rigorous statement is "I showed up more or less often". Everything else is interpretation and deserves that label.

The Discover report: a different panel with different rules

It is a separate report for AI features inside Discover. Two things change. One, an impression counts when the link scrolled into view, card or carousel, and once per result and session. Two, here everything aggregates by page: two results from your site in the same list count as two impressions, the opposite of Search. Attribution is canonical again.

Why you cannot see the report (three real causes)

Not enough impressions: the help says so, a new or niche site can take months to cross the threshold. You excluded the site through the generative AI control, and without eligibility there is no data. Wrong property or wrong report: looking at a domain property while comparing URL prefixes, or hunting for the Discover one thinking it is Search. Add a nuance from the official AI optimization guide: the page must be indexed and eligible to be shown with a snippet; with a `noindex`, an empty report is a symptom, not the cause.

Four practical ways to read the data without fooling yourself

Cross AI impressions with web clicks

Same date range, Web: text filter on both sides; compare the AI impressions curve with clicks and CTR, page by page. Pattern to watch: AI impressions rising while clicks fall. It suggests the AI answer resolves the query without a click; it is not proof of causality.

Compare countries and devices before rewriting anything

AI Overviews and AI Mode availability is not uniform by country or device. A drop in one country may be rollout, not weak content. Change something only when the pattern repeats across countries and weeks.

Use top pages as a topic map, not a ranking

The pages with the most AI impressions are the ones Google considers useful sources. They tell you where to invest: refresh data, add tables, answer the question in the first paragraph. They do not tell you those pages get traffic; thousands of impressions with zero clicks can still be a branding win.

Annotate changes and stop confusing cause with coincidence

Use Search Console annotations to mark deploys and campaigns. Without them any AI chart is read three different ways. And with no clicks and no queries, correlation is the most you will get.

Impressions up, clicks flat: what to do

Three usual causes: the AI answer resolves the query without a click, your link sits in the collapsed part, or the query changed intent. What to do: strengthen the value of the click (what only exists behind the visit: data, tools, templates), review title and snippet on your highest AI-impression pages, and measure where business lives, in conversions, which ties into why Search Console and Analytics disagree.

What not to do: create one page per sub-query inside an AI answer. Google marks that as a violation of its scaled content abuse policy and describes it as ineffective long term. If you are coming from a general drop, start with search traffic drop in Google. The report tells you where you appear; the business is told by conversions.

A seven-point monthly checklist

1. Can I open the Search report? If not, is it the threshold or an exclusion through the AI control?
2. Am I comparing the same date range and search type, text and multimodal apart?
3. Did I use the chart for the trend and the table for comparison?
4. Did I discard preliminary data from the last few days?
5. Did I cross my ten highest AI-impression pages with their clicks and CTR?
6. Did I check countries and devices to rule out a rollout?
7. Did I annotate the site changes in that period?

That is what the report offers: knowing where you appear and comparing trends. What it does not offer is how much traffic you gain or lose.