How to Measure the Clicks AI Answers Take From Your Site

Why AI answers reduce clicks

When Google shows an AI Overview, the searcher often gets what they need from the summary and never reaches the links below it. Pew Research Center measured this directly with a consented browsing panel of 900 US adults and 68,879 searches from March 2025, published on 22 July 2025. Users clicked a result on 8% of visits that showed an AI summary and on 15% of visits that did not. Only 1% clicked a link inside the summary itself, and 26% ended their browsing session after a page with a summary, against 16% after a page without one.

Larger keyword studies point the same way. An SEO software company that sells keyword and traffic tools studied 300,000 keywords and reported in April 2025 that position 1 desktop CTR was 34.5% lower than forecast when an AI Overview appeared. Its February 2026 update on 300,000 keywords, comparing December 2023 with December 2025, put the drop at 58.0% for position 1, 50.8% for position 2 and 46.4% for position 3. The marketing agency Amsive studied 700,000 keywords across 10 sites in five industries (April 2025) and found non-branded CTR down 19.98% after a keyword began to trigger an AI Overview, while branded CTR rose 18.68%. For Google's conversational AI Mode, a clickstream study of 69M US desktop sessions from May to July 2025, published by another SEO software vendor, found that 92% to 94% of AI Mode sessions had no external click.

Every independent study agrees on the direction, and the loss concentrates in non-branded, informational queries. Google has said that total organic click volume is "relatively stable" year over year (Liz Reid, August 2025), and it has published no figures to support or size that claim.

Two kinds of answer: modeled and observed

There are two honest ways to put a number on this for one site, and they answer different questions. The first is a modeled estimate built from public data. It starts from the keywords a site ranks for on page one, notes which of them show an AI Overview, and applies published click penalties to the traffic those positions would normally earn. It needs no access to the site's own data, which makes it useful for early planning.

A modeled estimate carries the weaknesses of its inputs. Keyword research tools miss much of the long tail, and AI Overview presence changes from day to day and by device and location. The traffic figures these tools report come from search volume multiplied by an average click rate for each position, so they apply no AI penalty at all. A page that holds position 1 under an AI Overview keeps the same estimated traffic in the tool after most of its clicks are gone.

The second answer is an observed measurement taken in the site's own Google Search Console. Search Console reports the clicks and impressions Google actually recorded for the site's queries, so it describes what happened on that site. When both exist, the observed result is the stronger evidence. A careful report labels which kind of figure it presents, gives the date, and names the source behind every input.

Why clean date windows matter

Search Console data has its own history of distortions, and a comparison that crosses one of them can produce a false result. Google's Data Anomalies page states that a logging error over-reported impressions from 13 May 2025 until a fix in late April 2026, and that clicks were not affected. Separately, around 10 September 2025 Google stopped honoring the num=100 parameter, which loaded 100 results per page. In data from 319 properties reported by Search Engine Land, 87.7% lost impressions after that change, while clicks barely moved and average position appeared to improve.

These events matter because the familiar picture of AI click loss is rising impressions alongside falling clicks. Part of that rising line was a reporting error. Clicks are the steadiest series for any trend that crosses these dates, and impressions and CTR are only comparable when both periods sit on the same side of each problem. Search Console also keeps only 16 months of data, so an older baseline depends on an export someone saved earlier.

Core algorithm updates add a further hazard, because rankings move while an update rolls out, as they did from 27 March to 8 April and from 21 May to 2 June 2026. A fair comparison keeps those weeks out of the periods it compares.

Why a comparison group matters

A site can lose clicks for many reasons at once, including seasonality, ranking changes, a migration and broad shifts in search behavior. Looking at AI-affected queries alone cannot separate these causes. The remedy is to compare two groups of the site's own queries over the same windows: queries that show an AI answer, and similar queries that do not.

The comparison group absorbs everything the two groups share. If both fall by the same amount, the cause lies somewhere other than AI answers. If the AI group falls further while the comparison group holds, the gap between them is the best available estimate of the AI effect. The background trend can be large: a study by a search marketing agency that sells related services, covering 3,119 informational queries across 42 organizations and 25.1M impressions, published in November 2025, found organic CTR on AI Overview queries fell from 1.76% to 0.61%, and queries without an AI Overview also fell 41%. Without subtracting that background, a report would blame AI answers for the whole decline.

It also helps to hold ranking steady, since a query that dropped several positions lost clicks for a ranking reason. Branded queries belong outside both groups, because branded clicks behave differently and can rise when an AI Overview appears.

How to read the result honestly

Each cause leaves its own signature in the data. When AI answers are the cause, clicks fall in the AI group at a stable average position, impressions hold steady or rise, and the comparison group stays flat or follows its usual season. If average position falls along with clicks, the cause is a ranking loss. If impressions drop on one date in both groups, a reporting change is the likelier explanation, and if both groups fall in the same weeks they fell last year, the pattern is seasonal.

A sound result states its limits. It names the windows, the events inside them, whether the figure is modeled or observed, and its date, because the effect is still moving. The same agency's 2026 update, covering 53 brands, 5.47M queries and 2.43B impressions from January 2025 to February 2026, saw CTR on AI Overview queries recover from 1.31% in December 2025 to 2.36% in February 2026.

Modeled estimates have known biases in both directions. They understate the loss where AI Mode use is high, since AI Mode cannot be observed keyword by keyword, and they overstate it for strong brands and for pages cited inside the answer. Google's Search Generative AI performance reports, which began on 18 May 2026 in a limited rollout, show impressions in AI features by page, country and device. They report no clicks and no queries.

The other half of the journey

Click loss covers only part of what AI answers do to a buyer's path. A buyer who reads an answer that names a brand often acts later, by searching for that brand, typing its address, or arriving from the AI assistant itself. Branded search, visits referred by AI assistants, and conversions from branded and direct traffic are where that value shows up. Growth in branded search while non-branded clicks fall is the pattern this second-search behavior predicts.

What this means for brands

Treat every AI click-loss figure as a dated estimate with a stated basis. A modeled range from public data is a reasonable starting point for planning, and an observed result from the site's own Search Console, built on clean windows and a comparison group, is the figure to act on.

Being named inside the answer changes the outcome. The agency's 2026 study found that brands cited in an AI Overview kept a CTR of 2.07% against 0.94% for brands left out, across 53 brands. The practical goal is a place among the sources the answer cites, followed by measurement of the searches and visits that come after.

Sources

• Google users are less likely to click on links when an AI summary appears in the results, Pew Research Center, 22 July 2025
• Google AI Overviews study, Amsive, 16 April 2025
• AI in Search is driving more queries and higher quality clicks, Google, Liz Reid, 6 August 2025
• Data anomalies in Search Console, Google Search Console Help, entry dated 3 April 2026
• Google num=100 impact data, Search Engine Land
• Google modifies search results parameter, affecting SEO tools, Search Engine Journal, September 2025
• Google Search Status Dashboard, ranking incident history, Google
• Search Generative AI performance reports, Google Search Central Blog, June 2026