How to Measure AI Overview Clicks Search Console Won't Show

What Google gives you, and what it holds back

Start with what exists, because it is more than most teams realize. Google's help documentation says the generative AI performance report was rolled out to all websites worldwide as of August 31, 2026. It shows how often links to your site appeared in AI Overviews and AI Mode, with four dimensions: pages, countries, devices and dates. Search Engine Roundtable reported when it first appeared in the UK that the data starts on May 18, 2026.

What it leaves out is everything you would use to judge value. There are no clicks, no click-through rate and no queries. Asked about click data, Gary Illyes wrote on November 2:

The team is super interested in giving more data, data that makes sense, but we don't have anything to announce yet.

The clicks are not missing from Search Console. They are mixed in. Google's AI features documentation says sites appearing in AI Overviews and AI Mode are reported in the Performance report within the "Web" search type. The counting rules add three details that matter for any analysis:

• Clicking a link to an external page in an AI Overview or AI Mode counts as a click.
• An AI Overview occupies a single position, and every link inside it is assigned that same position.
• A follow-up question in AI Mode is counted as a new query, with its own impressions and clicks.

That second rule explains a pattern many teams have misread. A page cited in an AI Overview at the top of the results can show an average position near 1 with a click-through rate that looks broken. The position is real. The click opportunity it used to represent is not.

Where each AI surface shows up in your data

Before you measure anything, know where each source lands. Most reporting errors come from looking for AI traffic in the wrong place.

The GA4 column comes from Google Analytics' default channel definitions. The AI Assistant channel covers sources "like ChatGPT, Gemini, Deepseek, Copilot, or Grok," and the documentation states plainly that it "excludes Google's AI Overviews and AI Mode." Google's AI answers arrive as Google organic traffic, which is why no analytics setting will ever isolate them.

Method one: calculate an AI exposure ratio for every page

This is the method to build first, because it uses only Google's own data and works at page level. For each page, divide its impressions in the generative AI report by its impressions in the regular Performance report, using the same date range, country and device filters. The result is the share of that page's search visibility that happened inside an AI feature.

Then sort. A page with a 60% exposure ratio and a click-through rate that dropped by half since last year is a strong candidate for AI absorption. A page with a 5% ratio and the same drop has a different problem, usually rankings, intent or a SERP feature that is not AI.

Here is how the arithmetic works on an illustrative page. Say the Performance report shows 40,000 impressions and 600 clicks for a month, a 1.5% click-through rate. The generative AI report shows 24,000 AI impressions for the same page and period, so the exposure ratio is 60%. If the same page earned 3% on similar impressions before AI Overviews covered its queries, you have a defensible estimate that AI exposure is costing it roughly half its clicks.

Method two: build a query baseline before and after

The exposure ratio tells you where. A query baseline tells you how much. In the Performance report, export query and page data for a period before AI Overviews were common on your queries, and a recent period of the same length. Keep only queries where your average position stayed roughly stable, within one position. If position held and click-through rate fell, something on the page took the click.

AI Overviews are the main suspect on most informational queries. Search Engine Roundtable reported on October 26 that they now appear on more than 80% of Google searches, citing data from Semrush, SE Ranking and seoClarity, three companies that sell SEO tools. But they are not the only suspect. Rankings were volatile through October, with unconfirmed update spikes on October 24 and 27, so check your date ranges against known volatility before blaming AI.

Method three: split branded from non-branded and by intent

AI answers do not hit every query equally. A searcher typing your brand name usually wants your site, and AI summaries rarely satisfy that intent. A searcher asking how something works is often satisfied by the summary. Filter the Performance report by a regex of your brand terms, then compare click-through rate trends for branded and non-branded queries separately.

Then go one level further on non-branded queries. Question-style and long queries are the ones most likely to trigger AI summaries. Pew Research found that 8% of one- or two-word Google searches produced an AI summary in March 2025, against 53% of searches with ten words or more. If your long, question-style queries lost clicks while short commercial ones held, you have a strong signal pointing at AI answers rather than at your content.

Method four: track which queries show AI answers and cite you

Search Console tells you that a page appeared in an AI feature. It does not tell you which query triggered it. A rank tracker that records AI Overview presence and citations for a fixed keyword set fills that gap. Track the queries that matter commercially, record weekly whether an AI Overview appeared and whether your domain was cited, and join that to Search Console clicks for the same queries.

Third-party tracking has its own limits. Search Engine Roundtable reported in October that rank trackers are still struggling after Google's anti-scraping changes, and Google may be testing the removal of citation cards beside AI Overviews, so treat tracker data as a sample, not a census. A few hundred well-chosen queries tracked consistently will tell you more than thousands tracked badly.

Method five: measure AI assistants with GA4 and server logs

ChatGPT, Gemini and Copilot are a separate problem, because their traffic leaves Google entirely. Start with GA4's AI Assistant channel, which assigns the medium "ai-assistant" when the referrer matches Google's list of AI assistants, according to the channel definitions.

Then account for what it misses. Search Engine Journal reported that ChatGPT adds UTM parameters to links in answers grounded in web sources. Gemini added utm_source=gemini in early September, then, per Search Engine Roundtable, stopped around September 28. Gemini on the desktop web still passes a referrer. Its iOS app opens links in its own WebView and sends none, so those visits land in Direct. Watch for a rise in Direct traffic to deep content pages that nobody types by hand: that is often unattributed AI traffic.

Server logs add a third view. OpenAI's crawler documentation says ChatGPT-User visits a page when a user asks ChatGPT a question and it fetches that page, while OAI-SearchBot crawls for ChatGPT's search features. A ChatGPT-User request is not a visit, but a page that ChatGPT fetches often is a page it is using to answer questions. Count those requests by URL each month.

For Microsoft, Bing Webmaster Tools offers what Google does not. Its AI Performance report, public since February, shows citations across Copilot and Bing's AI summaries, cited pages and sampled grounding queries. It still has no clicks, but grounding queries are the closest thing to AI query data any engine provides.

Method six: the opt-out test, if you can afford it

The cleanest measurement is also the most expensive. Search Console's Search generative AI control lets a site opt out of AI Overviews and AI Mode, and Google says opted-out sites will not receive traffic or impressions from those features, per Search Engine Roundtable's coverage of the launch.

One public case study shows both the value and the risk. Cyrus Shepard, who runs an SEO consultancy, opted a site out and compared Search Console data before, during and after, as reported by Search Engine Roundtable on October 30. He estimated AI impressions earned about a 0.9% click-through rate. Regular organic clicks also fell while the site was opted out, which suggests AI citations may support clicks elsewhere on the page.

What the independent numbers say, and how to use them

Borrowed benchmarks are useful for sanity checks and dangerous as forecasts. The strongest independent one is still Pew Research Center's analysis of 68,879 Google searches by 900 U.S. adults in March 2025. Users clicked a traditional result on 8% of visits when an AI summary appeared, versus 15% when none did. They clicked a link inside the summary on 1% of visits.

The largest first-party disclosure came last month. Reddit's Q3 2026 shareholder letter said search traffic dropped by more than 1 million referrals per week, on average, during October, and its CEO named AI Overviews and AI Mode as the biggest part of the company's working theory. Google sees it differently: its documentation says clicks from results pages with AI Overviews are "higher quality," meaning users are more likely to spend more time on the site.

Both can be true, which is why your own measurement matters. If you track fewer clicks, check engagement and conversion rate on the clicks you keep. A page that lost 40% of its clicks but kept its conversions has a reporting problem, not a revenue problem.

For how to present this to leadership, our guide to SEO reporting for executives covers framing, and our Search Console strategy guide covers the exports these methods depend on. For how AI Overviews have changed through the year, see our running notes on AI Overviews updates in 2026.

Where to start

Google has decided, for now, that you can see where AI uses your pages but not what that use is worth. That is a reporting gap, not a dead end. Teams that build their own estimate this quarter will be arguing from evidence when the next budget conversation starts, and everyone else will be arguing from screenshots.