How to Track Google AI Overviews: Presence, Mentions and Links

What Tracking an AI Overview Measures

Every check of an AI Overview answers three separate questions, and each needs its denominator stated:

• Presence rate: the searches that showed an overview, out of all the searches you tracked.
• Mention rate: the searches whose overview named you, out of all the searches you tracked. The share of overviews that named you is a useful second view, as long as it is labelled as one.
• Link rate: the searches whose overview linked a page on your site, out of all the searches you tracked.

Search Console adds a fourth figure, impressions of your links, which counts times shown rather than giving a rate. None of these is a rank. Search Console Help says an AI Overview occupies a single position in search results and all of its links share it, so an “AI Overview rank” is a figure a tool works out for itself, usually from the order of the reference cards. Google's own guide adds that no third-party tool has access to its internal ranking or AI systems.

Start With Search Console's Generative AI Report

Every site now has a free census of its own links in Google's AI features. Google's Search generative AI performance report, announced on 3 June 2026 and rolled out to all sites by 31 August 2026, shows how often URLs from your site appeared in AI Overviews and AI Mode, by page, country, device and date (Discover's AI features have a report of their own). It counts an impression each time a link to your site was shown in one of those features.

It is a census of your own links and nothing else. Google lists impressions and no query dimension, so the report can't tell you which questions brought the overview up, whether an overview named you without a link, who else it cited, or how AI Overviews compare with AI Mode, which it reports together. A site opted out with the Search generative AI control gets no AI feature impressions at all. And Google Analytics won't separate the visits: GA4's AI Assistant channel leaves out AI Overviews and AI Mode, so a click from an overview counts as Organic Search. Use the report as the baseline your other numbers should agree with.

How We Tested

To measure what Search Console leaves out, we ran a test of our own:

How Often Google Shows an AI Overview

Across all 1,200 questions, 92.8% of results pages showed an AI Overview. The shape of the question mattered most. Comparisons and problem questions showed one on 100% of searches and how-to questions on 98.6%, while the two most commercial shapes showed one least often: “best” questions on 83.3% and short keywords on 81.9%. Markets differed too: 94.7% of searches in India showed one, against 89.4% in the United Kingdom, where short keywords showed one on only 73.6%.

We picked questions buyers ask, so these rates run well above the web as a whole. Pew Research Center, studying the Google searches of 900 US adults in March 2025, found an AI summary on 18% of searches: 8% of one- or two-word searches and 53% of searches of ten words or more. An academic study for SIGIR 2026 found overviews on 51.5% of a set of real user queries. Your presence rate is partly a property of the questions you chose, so track the questions your buyers ask, and don't compare your rate with a rate measured on someone else's questions.

Keep the Searches With No Overview in the Count

A rate counted only over the overviews you saw answers a fair question, how you compete inside an overview, but it moves when Google changes how often it shows them, whether or not anything changed for you. Counted over every search you tracked, the same brand reads lower, and the number describes what a buyer actually sees.

There is a second way to lose the denominator: a capture that misses the overview. In our test, 48.8% of the overviews loaded only after the rest of the page, and an overview can also sit collapsed behind a “Show more” button. A script that reads the page before the overview arrives records “no overview”, and one that reads only the collapsed teaser records part of it. When you choose a tracker, ask how it handles both.

Being Named Is Not Being Linked

An overview can recommend a brand without linking to it, and link to a page that recommends someone else. In our test, 89.2% of overviews named at least one brand, 4.6 on average, and linked 8.6 reference cards. Of every brand named, 71% had no link to its own site in that overview, and only 13.5% of the reference cards went to the own site of a brand the overview named. The most commercial shapes named the most brands: 6.4 on average for short keywords and 6.2 for “best” questions, against 2.4 for problem questions.

So track mentions and links in separate columns. A brand that is named but not linked gets the recommendation and loses the click. A brand that is linked but not named supplied a fact for somebody else's recommendation. The sites the overview does link are mostly other people's, and video leads them: youtube.com held 22.5% of the reference cards and appeared in 79.5% of overviews, and reddit.com appeared in 33.4%. Those sites are your off-site list, the layer covered in our GEO checklist.

Check More Than Once

The overview for a question is not fixed. In our test, 83.4% of question and market pairs showed an overview in all three searches, an hour apart, 1.2% in none, and the other 15.4% in one or two. The ones that came and went were the most commercial questions: short keywords showed an overview in all three searches for only 65.3% of pairs and “best” questions for 63.9%, while every comparison and problem question showed one each time. When a question showed an overview in two of its searches, the two overviews had 63.2% of their cited sites in common on average, counted over every site either one cited. Where all three searches showed an overview, only 48.5% of the brands those overviews named were named in all three. The same question asked in two different markets had only 24.4% of its cited sites in common.

Independent research finds the same. The SIGIR 2026 study found that two runs of the same query shared AI Overview sources at a Jaccard similarity of 0.66, against 0.78 for the classic results, and 0.64 across locations and 0.55 across devices. So fetch each question several times per measurement, report rates over a window, and compare windows, never single checks. We cover the same effect across AI engines in why you can't measure AI visibility once.

AI Mode Needs Its Own Measurement

AI Overviews and AI Mode share a search box, but Google says they may use different models and techniques, so the responses and links they show vary. Search Console counts them differently too: an AI Overview takes one position on the results page, while a follow-up question in AI Mode counts as a new query. Alphabet now describes the two as one Search experience, and Google counts more than 2.5 billion monthly users of AI Overviews and more than a billion of AI Mode. AI Mode can also personalise answers from Search history for users aged 18 or over who have Search history and personalised recommendations turned on. A results-page tracker reads AI Overviews; AI Mode needs its own measurement, and a signed-out check can't show what a signed-in user sees there.

How to Track AI Overviews by Hand

1. Freeze your questions. Pick the buyer questions that matter, mostly unbranded and written the way people search, and change the list only with a dated note. Our guide to choosing prompts to track covers which ones can show a win.
2. Pin the market, language and device for each question, and use a clean, signed-out browser.
3. Fetch each question at least three times per measurement, spaced out.
4. Wait for the overview and expand it before you record anything.
5. Record one row per fetch with the columns below.
6. Report presence, mention and link rates over every fetch, and compare this window with the last one.

The limits are time and reach. Forty questions fetched three times in three markets is 360 rows every time you measure, and a signed-out browser in one place is one view of a page that changes with location, device and time.

Overview Gone, or You Gone? A Diagnostic

When your AI Overview numbers drop, work through these in order:

1. Did presence drop? If fewer of your searches show an overview at all, Google changed, not your page. Check the Search Console report for the same dates.
2. Did your mention or link rate drop inside the overviews that still show? Then look at what the overviews cite instead of you.
3. Is the page still indexed and snippet-eligible? Google says a supporting link must be indexed and eligible to show with a snippet. Run URL Inspection.
4. Did a template change add a snippet control? nosnippet stops content being used as a direct input for AI Overviews and AI Mode, and max-snippet limits how much can be used.
5. Is the Search generative AI control set to exclude? An excluded site leaves AI Overviews, AI Mode and Discover's AI features, usually within a few days of the change.
6. Did the cited sites change? Compare this window's linked sites with the last one's. A new review or publisher page in the overview usually explains a new brand in it.

How Ranqo Fits

Google AI Overviews is one of the six engines Ranqo tracks, on the Pro and Business plans and in the engine pool agency plans choose from. For each of your questions, Ranqo fetches the live results page in the question's target country, waits for an overview that loads late and expands a collapsed one, and records “no overview shown” as an answer that does not name you, so the searches with no overview stay in the count. The same questions run on the other engines your plan tracks in the same weekly run, with every number reported per engine.

The Sources page shows which sites the overviews cite for your questions, labelled by source type and page format, and the Action Center turns the gaps into ranked fixes whose lift is counted over the runs that follow.