How much traffic are AI Overviews costing you? An estimate you can run in Search Console

Why industry averages will not answer this for you

Industry studies show which way AI Overviews move clicks, not how far they move yours. Your queries, positions and page types differ from any study's sample, so the estimate has to come from your own data.

The studies agree on direction. In Ahrefs' analysis of 300,000 keywords, comparing desktop click-through rates in December 2023 and December 2025, the top-ranking page's click-through rate was 58% lower when an AI Overview appeared.¹ Ahrefs sells SEO software, so read it as a vendor study with a large sample.

Being in the Overview changes the picture. In Seer Interactive's data for 53 brands from January 2025 to February 2026, pages cited in an AI Overview earned about 120% more organic clicks per impression than when they were not cited.² That is a correlation, and cited pages may differ in other ways. Seer is an agency that sells search services.

Agarwal and Sen took a different route: a pre-registered randomized field experiment that hid AI Overviews from some US desktop Chrome users and compared outbound clicks and reported satisfaction.³ Randomization is the cleanest way to separate the Overview from everything else, and the working paper is worth reading for its design.

AI Overviews and clicks

Same direction, different sizes

58% lowerCTR for the top-ranking page when an Overview appearsAhrefs, Dec 2023 vs Dec 2025 ¹

About 120% moreclicks per impression when a page is cited in the OverviewSeer Interactive, Jan 2025–Feb 2026 ²

What do you need before you start?

You need Search Console data from before and after Overviews reached your queries, and a way to tell which queries show an Overview. The second part takes the most work.

Search Console's AI reports, rolled out worldwide on 31 August 2026 according to Search Engine Journal, show impressions in AI Overviews and AI Mode by page, country and date.⁴ They do not show clicks yet. They tell you which pages appear in Overviews, which is your list of cited pages, but not a click-through rate. Our note on the rollout covers what the reports count.

For the query split, use a rank tracker that records search features, or check your top queries by hand in a clean browser in the markets you serve. Record each query as "Overview, your page cited", "Overview, not cited" or "no Overview". Overviews come and go, so check each query on more than one day.

The estimate, step by step

Compare how click-through rate changed for Overview queries with how it changed for queries without one, at similar positions. The difference between those two changes is your estimate of the Overview's effect.

Pick two comparable periods

Use the same months a year apart if your data reaches back that far, so seasonality cancels out. Otherwise, compare the weeks before and after Overviews first appeared for your main queries.

Split the queries into three sets

Export query-level data from the Performance report with clicks, impressions and average position. Tag each query with its set from your feature check, and leave branded queries out for now, because they behave differently.

Compare at like positions

Group queries by position: the top result, positions two to three, and four to ten. Within each group, calculate click-through rate for each set in each period. The no-Overview set is your control. Its change captures seasonality, demand and site-wide ranking shifts, so subtract it from the change in the Overview sets.

Turn the gap into clicks

For each Overview set, take its earlier click-through rate, adjust it by the control set's change, and compare the result with the actual rate. Multiply the gap by current impressions to get clicks lost per month.

Illustrative. Contoso Advisory is a fictional consulting firm and every number is invented. The control fell by a tenth, so expected rates are 27 per 100. With 20,000 monthly impressions on uncited Overview queries, the gap of 12 per 100 is about 2,400 lost clicks a month; with 10,000 on cited queries, the gap of 3 is about 300.

The estimate in five steps

1. Periods

   Same months a year apart, or before and after Overviews arrived.

2. Three sets

   Overview and cited, Overview and not cited, no Overview.

3. Like positions

   Click-through rate by position group, per set and period.

4. Clicks lost

   Expected minus actual rate, times current impressions.

5. Range

   Low and high from the spread across groups and periods.

Why report a range, not one number?

Because the method has error at every step, and a single figure hides it. Take the lowest and highest estimates across position groups and periods, and report both with the method beside them.

Three things distort the result. Rank changes that coincide with Overviews look like Overview losses, which is why the comparison runs at like positions. The query mix shifts between periods, so keep the same query list in both. And Overviews appear on some days and not others, so your feature check is a sample, not a census.

Country matters too. Search Console mixes every market by default, while Overviews differ by country and language. Filter to the countries you sell in and run the estimate for each one that carries real volume.

Behavioural research suggests the effect is real, not an artefact of method. In Pew Research Center's browsing data from 900 US adults in March 2025, now more than a year old, people clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when none did.⁵ The size on your site is still yours to measure. Our guide to tracking AI Overviews and AI Mode separately explains why to keep the two surfaces apart.

What will make this easier?

Better referral data is arriving, but slowly and without documentation. Search Engine Journal reported on 1 October 2026 that Gemini now adds UTM parameters to outgoing links, a change Google has not documented.⁶ If it holds, Gemini visits become easier to separate, and our GA4 AI referral channel guide shows where to catch them.

Until Search Console reports clicks from AI Overviews, an estimate like this is the closest you will get. Our view: put the range and the method in the report, and decide with the board which single number to track, as our board metrics guide describes.

Lost clicks are only half of it; whether the Overview names you is the other half. Sigzen AI's audit measures that across four engines (ChatGPT, Gemini, Google AI Overviews and Google AI Mode), running each prompt on three separate days through a licensed AI-answer data provider. See what the audit covers.