How to track AI Overviews

Google counts AI Overviews in two places in Search Console, and the two answer different questions.

The Generative AI performance report shows impressions of your pages in AI Overviews and AI Mode. It groups them by page, country, date and device, separates text searches from multimodal ones such as Google Lens, and exports to a file; Google has rolled it out to every site since 31 August 2026^[1]. Its pages are the final addresses after any redirect, assigned in most cases to the canonical page, and the usual limit of 1,000 table rows applies^[1].

The Performance report counts the same traffic within the Web search type, together with the rest of the results page^[2]. There an AI Overview occupies a single position, every link in it is given that same position, and a click on one of its links to your site counts as a click^[3].

Two details matter when you compare periods. A follow-up question in AI Mode counts as a new query, and experiments in Search Labs are left out of the data^[3]. So Google tells you how often and where your pages were shown, but not on which queries an overview appeared, what it said or whether it gave you an AI citation. That part you track yourself.

A page can appear in an AI Overview only if it is indexed and eligible to be shown in Google Search with a snippet^[2]; the full list of what keeps a page retrievable is the first step of the hub’s practical guide, and the audit checklist turns it into tests. One setting is particular to Google’s AI features, and it is worth checking first.

1. 1Open Settings, Search generative AI in Search Console. Including the site is the default, and a property inherits the choice of its closest parent property that set one^[4].
2. 2Check every property you own, domain and URL-prefix alike: a parent set to exclude takes its child properties out with it, unless a child sets its own choice^[4].
3. 3If a site is excluded, its links and content leave AI Overviews, AI Mode and Discover’s AI features, generally within a few days; content from other sites stays and may look similar to yours^[4]. Excluding a site is a way out of the feature, not a way to protect the topic.
4. 4Keep training apart from appearance. Google says this control does not affect AI training, and points to Google-Extended to limit the training of the models that write these responses^[4].

No special file or markup is needed: Google says you need no new machine-readable files, AI text files or special schema.org structured data to appear^[2]. Whether an llms.txt is worth publishing for other services is a separate question, with a guide of its own.

An AI Overview appears only when Google’s systems judge that it adds something to the classic results, and on many queries it does not appear at all^[2]. A list that records the overview’s absence as well as its presence shows where the feature spreads, not only where you are cited.

Build the list from your own Search Console queries rather than from a keyword tool. Four kinds of query earn a place on it:

• Questions your pages already rank for, especially complex how and why questions: Google describes AI Overviews as a way to get the gist of a complicated topic or question quickly^[2].
• Queries that bring revenue or leads, even where no overview appears today: you want to know the day one does.
• The obvious follow-up questions of each. AI Overviews and AI Mode may fan a query out into several related searches across subtopics^[2], so a page that also answers the next question is useful for more of them.
• A few queries where a competitor is cited and you are not, as a control for your own changes.

Keep the list fixed for each period, and record country, language and device with every check, since the report itself splits impressions along those lines.

One row per query and check is enough. What matters is that every row is recorded the same way, whether a person or a tool does the checking:

date,query,country,device,overview,our_url,link_order,others_cited,organic_position2026-10-06,how to descale a kettle,GB,mobile,yes,https://shop.example/descale,2,brand.example;forum.example,42026-10-06,best kettle for hard water,GB,mobile,yes,,,review.example;brand.example,72026-10-06,kettle warranty claim,GB,mobile,no,,,,1

Two fields need a written rule before the first check. link_order is the place of your link among the overview’s sources as they are listed, counted from the top. organic_position is your best classic result on the same page. Keeping both shows whether the overview cites you because you rank well, or despite a low rank.

How often to check depends on the list: weekly for the short list of queries that bring revenue, monthly for the rest. An overview can change between checks, so one absence is a reading, not a result. For the questions people ask assistants rather than Google, the same discipline applies in how to track your brand in ChatGPT.

A rank tracker that records AI Overviews saves the manual work, and it is only as good as the conditions it checks under. Before trusting its figures, ask five questions:

1. 1From which country, language and device does it check, and can it match the markets you sell in?
2. 2How often does it check each query, and does it record a check where no overview appeared?
3. 3Does it keep the full text of the overview and the list of its sources, or only a yes or no for your domain?
4. 4Does it record the order of the sources, so that placement can be read and not only presence?
5. 5Can you export the raw rows, so that its figures can be checked against your own log and against Search Console?

A tool that cannot answer the third question reports a score you cannot check against the answer it came from. Keep a short manual log beside it for the queries that matter most, and compare the two each month.

Search Console can show the effect of AI Overviews on a page even where it cannot name them. Compare two periods for the pages you care about:

1. 1In the Performance report, choose Compare in the date filter and pick two periods; the table then shows a Difference column, and weekly or monthly granularity removes the day-of-the-week effect^[5].
2. 2Filter to the pages you track and read impressions, clicks and position together for each.
3. 3Mark the pages whose impressions hold while clicks fall, on informational queries above all.
4. 4Check those pages’ queries in your log, or by hand, for an AI Overview.
5. 5Open the Generative AI performance report for the same pages and period, to see whether their impressions in AI features rose.

Position needs care in this reading. Because every link in an AI Overview takes the overview’s position^[3], a page cited in an overview at the top of the results can report a high position and few clicks at once. That is the overview at work, not a ranking problem.

The two Search Console reports can be joined. The Generative AI performance report draws its data from the Web search type of the Performance report^[1], and both assign most data to the canonical page^[1][3]. Export both for the same period and pages, and one division shows how much of a page’s visibility now comes through AI features:

ai_share = impressions in the Generative AI report           ÷ impressions in the Performance report, Web # Sample data: /guides/descale, September# 3,100 AI-feature impressions ÷ 12,400 Web impressions = 0.25

1. 1Use the same dates, the same search type and pages grouped by page in both exports, so that the two counts are aggregated the same way.
2. 2Compute the share for each page you track, then sort: the pages with the highest share are the ones whose clicks depend most on how AI Overviews and AI Mode treat them.
3. 3Track the share month by month beside the page’s clicks. A rising share with falling clicks is the overview taking the click; a rising share with steady clicks is reach you did not have before.
4. 4Read a page with few impressions as a note, not a figure: a share of a small count moves a long way on a handful of searches.

Visits from AI Overviews and AI Mode reach Google Analytics 4 inside Organic Search, not in its AI Assistant channel; how to count AI visits there, and which figures belong in a report, is set out in the metrics and KPIs that matter.