Search Console AI Overview Report: How Impressions Are Counted

TL;DR

• The Search Console AI Overview report, officially the "Generative AI performance report," launched on June 3, 2026 and opened to every site worldwide on August 31. It covers AI Overviews and AI Mode.
• The report has a single metric, impressions. It has no clicks, no CTR, no average position and no queries.
• An impression counts each time a link to your site is shown. Two links from the same site in one AI answer count once in the chart total, while the Discover version of the report counts them twice.
• Generative AI impressions are a subset of the regular Performance report, not additional traffic.
• AI Overviews and AI Mode are combined and cannot be viewed separately. In my test of 8 Traditional Chinese queries from Taiwan, about half (55%) of the domains AI Mode cited also appeared in the AI Overview for the same query.
• The Search Console API does not return this data. The only way to get it is a manual export, so building a time series means downloading and saving it yourself on a schedule.

Contents

Search Console AI Overview report coverage and dimensions
How AI Overview impressions are counted
Deduplication rules in the Search and Discover reports
The scrolled-into-view misconception for AI Overview impressions
Generative AI impressions and the Performance report
Generative AI data and the Search Console API
Page-level comparison of AI impressions and clicks
AI Mode vs AI Overviews citation sources
Current limitations of the generative AI report

Introduction

Google announced on the Search Central Blog on June 3, 2026 that Search Console was adding a report dedicated to impressions from generative AI features, the report most people call the AI Overview report. It started as a limited test and opened to all sites on August 31. Before that, a site's appearances in AI Overviews were mixed into the regular Performance report with no way to separate them.

Two misreadings have been common since the report opened. One treats AI impressions as a new source of traffic, and the other reads "impression" as "a user actually saw it." Neither matches Google's counting rules, and the details of those rules change how the numbers should be read.

This article walks through how the report counts, rule by rule, based on the Search Console help documentation. It also covers two things the documentation doesn't spell out but that you run into as soon as you work with the data, which are that the API can't return it and that page-level reconciliation with the Performance report never quite balances. The comparison data comes from a Traditional Chinese site I run and from queries asked in Taiwan, which is useful mainly because nearly all published research on AI search features is based on English queries in the US.

Search Console AI Overview report coverage and dimensions

The report covers two features, AI Overviews and AI Mode, according to the help article for the Search version. Discover has its own separate generative AI report, and neither report includes the other's data.

The Search version has an export button, and both the chart and the table can be downloaded. According to the launch post, dates can be broken down to the hour, and only the Search version offers the device dimension.

The report has no clicks, no click-through rate (clicks divided by impressions), no average position and no queries. Missing queries matters most, because the report can tell you which page appeared in an AI answer but not which questions made it appear. Google's launch post says it will work with site owners to add more metrics over time; for now, impressions are all there is.

How AI Overview impressions are counted

An impression counts a link being shown, because Google defines it this way: "Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search." The documentation doesn't say whether an AI answer that names your brand without linking to you counts. A literal reading of the definition says it doesn't, but that is an inference rather than a stated rule.

When the same site appears with several links in one AI answer, the documentation says: "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." The deduplication applies to the chart total and works at the site level. As written, the table grouped by page still counts each page separately, so the page-level impressions in the table can add up to more than the chart total.

A further rule comes from the general explanation of impressions, position and clicks: "An AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position." Because an AI Overview usually sits at the top of the results, a cited page shows a very high position in the regular Performance report while its CTR can be far below that of an ordinary result at the same position. This is one of the most common reasons a page shows a top-few position with a CTR close to zero.

Deduplication rules in the Search and Discover reports

The two reports handle multiple links from the same source in opposite ways. The Discover help article says: "if two generative AI results from the same property appear in the same Discover list, each impression is counted separately."

This difference creates a systematic gap when you compare the two reports, since the same exposure produces a higher impression count in Discover. The two reports shouldn't be added together or compared against each other.

Discover also has a scrolling rule, under which the same card scrolled past repeatedly in one session counts as one impression. It is easy to confuse with the Search version's same-site deduplication, but they rest on different things, and the Search rule is about the same site, not about scrolling.

The scrolled-into-view misconception for AI Overview impressions

A claim you often hear is that a link inside an AI Overview only counts as an impression once it is scrolled into view, or once the user clicks "Show more."

The general documentation's rule for ordinary results is that "an impression is counted whenever an item appears in the current page of results, whether or not the item is scrolled into view." Being on the current results page is enough. The scroll-into-view condition applies to carousels, and the expand condition applies to expandable elements such as FAQ rich results and "People also ask."

For AI Overviews, Google only says "Standard impression rules apply" and defines no separate scrolling or expanding condition. The "must be scrolled into view" idea is extrapolated from the carousel and expandable-element rules and isn't a rule written for AI Overviews. Under the standard rules, the AI impression count means the number of times a link appeared on a results page, not the number of times a user actually looked at it, so treating it as reach overstates real visibility.

Generative AI impressions and the Performance report

Generative AI data is already part of the regular Performance report, as the documentation states: "The generative AI performance report includes data from the Web search type in the Performance report (Search results)." It is a subset of the Web search type.

On a Chinese-language site I own, the total impressions pulled through the Search Console API for July 24 to August 20, 2026 were 36,284, exactly matching the Search Console interface, while AI impressions for the same period were 4,946, less than the total. If AI impressions were added on top, those two totals would not have matched.

Growth in AI impressions therefore doesn't mean the site gained new traffic. It means part of the existing impressions are now shown as AI answers. Those impressions usually have a lower CTR, so as their share rises, overall CTR gets diluted even when the content itself is performing no worse. A separate page-by-page study across 181 pages looks at how the AI share actually affects clicks: Does AI Overview eat your site traffic? (in Chinese).

Generative AI data and the Search Console API

The export button in Search Console is currently the only way to get this data.

In the Search Analytics query method of the Search Console API, the search type parameter accepts only six values: web, image, video, news, googleNews and discover. None corresponds to AI Overviews or AI Mode. In tests in August 2026, setting the parameter to aiMode, aiOverview or generativeAi always returned an error. Grouping by the search appearance dimension instead made the request succeed, but the response contained no rows.

Two kinds of users feel this most:

• Sites that need long-term tracking: Search Console keeps performance data for a limited time (16 months for the regular Performance report), and older data drops off. The only way to keep a time series of AI impressions is to download it manually on a schedule, for example exporting last month's data at the start of each month.
• Teams using third-party dashboards or automated reports: these tools mostly pull data through the API, so they can't get generative AI data until the API supports it. If a report shows an "AI impressions" column, check where it comes from and how it is calculated.

Page-level comparison of AI impressions and clicks

Judging whether AI impressions affect clicks requires matching the generative AI report against the regular Performance report page by page, because the AI report has no clicks. The steps are:

1. In the Generative AI performance report, pick a date range, group by Page and export the table to get AI impressions per page.
2. In the regular Performance report, choose the same date range and the Web search type, group by Page and export it to get each page's total impressions, clicks and average position.
3. Join the two tables on URL and calculate each page's AI share (AI impressions divided by total impressions).
4. Sort the pages with a high AI share by CTR and find those with a high AI share and close to zero clicks.

The join won't reconcile perfectly. Page-level data never quite matches the interface totals because of how search data is anonymized: some queries are left out of the detailed rows for privacy but still count in the totals. In my data, 12 pages also appeared in the AI report without appearing in the page-level Performance data for the same period. Gaps like these don't change the order of magnitude, but they make the data unsuitable for exact page-by-page accounting.

The pages that step 4 surfaces are usually definitions or explanations, which an AI can answer in a paragraph without the user needing to visit. Pages that offer a tool or require hands-on steps can keep a normal CTR even with a high AI share.

AI Mode vs AI Overviews citation sources

The Search report combines AI Overviews and AI Mode and offers no dimension to split them, so you can't tell which feature produced a page's AI impressions. The two are used differently, because AI Overviews appear at the top of ordinary results while AI Mode requires the user to switch into a conversational interface. If they cite very different sites, the combined number is hard to interpret.

From September 13 to October 12, 2026, I recorded the source links attached to AI Mode and AI Overview answers for 8 Traditional Chinese queries asked in Taiwan, on recommending web design agencies, recommending booking systems and building websites for local businesses. Both features were sampled once on the same dates, twice per query, for 16 AI Mode answers in total.

About half of the cited sources were shared, but the overlap varied a lot by query. For 美髮沙龍需要官網嗎 ("does a hair salon need a website"), all 4 domains AI Mode cited also appeared in the AI Overview, while for the Kaohsiung query only 1 of 9 matched. The combined report therefore blends two partly different sets of sources, and when a page's AI impressions rise you can't tell which feature cited it.

From a single site's point of view the results were more consistent. My two sites were cited in AI Overviews for 4 of the queries, and in 2 of those they were cited by AI Mode as well. There was no query where AI Mode cited them and AI Overviews didn't, so in this sample AI Mode brought no exposure beyond what AI Overviews already gave.

The sample is small. Each query was sampled only twice per feature, and the queries cluster around two topics, web design and booking systems, so the results shouldn't be generalized to other industries or other kinds of queries. AI answers can differ each time, and the overlap could change with more samples.

Current limitations of the generative AI report

• Impressions only, no clicks. A common assumption is that AI impressions bring no clicks at all, but links in AI answers can be clicked; the report just doesn't show the rate.
• AI Overviews and AI Mode can't be separated, since their impressions are combined.
• No queries. You can't see which questions triggered the AI answers and can only infer them from the queries the same URLs receive in the regular Performance report.
• No API access, so automated tracking has to wait for Google to open it.
• Google says it will keep adding metrics, so the counting rules and fields may change. Check whether the documentation has been updated before comparing across periods.

References

• Introducing Search Generative AI performance reports in Search Console, Google Search Central Blog, June 3, 2026
• Generative AI performance report (Search), Search Console Help
• Generative AI performance report (Discover), Search Console Help
• What are impressions, position, and clicks?, Search Console Help
• Search Analytics: query, Search Console API documentation