Clicks Fall From 15% to 8% When an AI Summary Appears: Measure Your Own Number

Google will not show you an AI Overviews line in Search Console. Its own documentation says traffic from AI Overviews and AI Mode is folded into overall search traffic under the Web search type, with no separate breakdown, so every dashboard promising you an "AI Overviews CTR" is inferring it. That does not leave you blind. Three numbers you already have access to, plus one you probably do not, will tell you within an hour whether AI answers are eating your clicks: the gap between impressions and clicks over matched periods in Search Console, referral sessions from assistant domains in GA4, and your AI crawler log. Here is the routine, the arithmetic for what to expect, and the controls Google and Cloudflare actually document.

The one dataset with published methodology

Most numbers circulating about AI search come from vendor studies that do not show their work. The exception is Pew Research Center's browsing-data study (published July 22, 2025, opened July 25, 2026). Pew tracked 900 US adults on its KnowledgePanel who shared browsing data during March 2025, covering 68,879 unique Google searches, 12,593 of which produced an AI summary, with results scraped April 7-17, 2025.

What it found, for the United States in March 2025:

About 18% of searches produced an AI summary, and 58% of respondents ran at least one such search that month.

Users clicked a traditional search result link on 8% of visits where an AI summary appeared, against 15% of visits where one did not.

Only 1% of visits included a click on a source link inside the summary itself.

Sessions ended entirely on 26% of pages with an AI summary, against 16% without.

The typical summary ran 67 words, and 88% cited three or more sources.

Two limits before you build anything on this. It is US data from March 2025, over a year old, and AI Mode has expanded considerably since. And it measures user visits, not per-query click-through rate for a given site, so it describes behaviour rather than your ranking. Google, for its part, states that clicks from result pages with AI Overviews "are higher quality (meaning, users are more likely to spend more time on the site)", a claim published on its AI features page without accompanying methodology. Fewer, better clicks is a plausible story. It is not a measured one, and neither is the alternative.

What the shape of the loss looks like in your numbers

This arithmetic is ours, and it produces an order of magnitude rather than a forecast. Take a site with 40,000 Google impressions a month and a 4% click-through rate, so 1,600 clicks. Apply Pew's relative gap: a drop from 15% to 8% is a 47% relative reduction in the chance of a click. If AI summaries appear on 18% of the searches that surface you, the expected loss is roughly 18% multiplied by 47%, about 8.5% of your clicks, or 136 clicks a month.

Now the reasons that number will be wrong for you, all of which point in the same direction. Query mix dominates: informational and question-shaped queries trigger summaries far more often than 18%, navigational and transactional queries far less. A site answering "how do I" questions is exposed several times more than a site ranking for "brand name pricing". And the 18% figure predates AI Mode's expansion. If your traffic is mostly explanatory content, treat 8.5% as the floor and go measure the real figure, which is the next section.

The four-step measurement routine

Search Console, impressions against clicks. Open the Performance report, set the date comparison to the last 3 months versus the same period last year, and read the two totals separately. The signature of answer-without-click is impressions flat or up while clicks fall, with average position unchanged. If position dropped too, you have an ordinary ranking problem, not an AI one. Google's own documentation for this report (opened July 25, 2026) lists search types of web, image, video and news, and documents no AI filter, so this differential is the measurement.

Do it again per query group. Filter queries containing "how", "what" and "why" and compare that group's click trend against your brand-name queries over the same window. If informational clicks fell while brand clicks held, you have isolated the effect to exactly the queries AI answers well.

GA4, referrals from assistants. In the traffic acquisition report, switch the dimension to Session source / medium and look for chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. This is the counter-flow: traffic arriving because an assistant cited you. It is usually small and usually growing, and it will not show up in any SEO tool.

Check who is crawling you. If your site is on Cloudflare, AI Crawl Control is available on all plans, including free, at dash.cloudflare.com under your account and zone. The Crawlers tab shows which AI services are fetching your pages and the Metrics tab breaks activity down by crawler, status code, hostname and path, though free plans are limited to a 24 hour analytics window and to user agent string detection rather than full bot management.

The controls, and what each one costs you

Google documents the levers on its AI features page. None of them is free.

Google states that eligibility to appear as a supporting link in AI Overviews or AI Mode requires only that a page be indexed and eligible to appear with a snippet, and that there are "no additional technical requirements". Read that in both directions: there is no schema trick that gets you cited, and there is no way to be in Search but out of AI Overviews without also degrading your ordinary listing.

A defensible decision rule: block a crawler only when the content is the product. If people pay for your writing, or you sell a database whose value is its completeness, blocking a crawler that never sends referrals is straightforward arithmetic. If your content exists to bring in customers, blocking removes the citation without restoring the click, and the citation is the only remaining distribution you get from that query.

What actually responds to effort

The queries that lose clicks to AI answers are the ones with short, complete answers: definitions, conversions, opening hours, syntax. The ones that hold are those where the click is the point, because the reader wants your data, your tool, your prices, your judgement, or to hire you. Rebalancing toward the second category is a slow move and the only durable one, and it is why we keep coming back to owning distribution rather than renting it. If you serve customers face to face, the profile that surfaces in the local pack is a separate and still-clickable surface worth setting up properly. And every channel discussed here routes through a company that can change the deal, which is the standing argument for a list you own.

Do the measurement before the strategy. An hour in Search Console and GA4 tells you whether you are losing 2% of clicks or 30%, and those two situations call for entirely different responses. Guessing calls for neither.