AI Overviews Cut Clicks 58%. What That Number Actually Measures

What was actually reported

Three separate measurements are usually quoted as if they were one, and they answer different questions.

Ahrefs, 4 February 2026. The study compared 150,000 keywords that trigger an AI Overview against 150,000 that do not, using aggregated Search Console data for desktop, December 2023 against December 2025. Position-one click-through rate on AI Overview keywords fell from 0.073 to 0.016. The forecast for where those keywords would have landed without AI Overviews was 0.037. That gap is the 58% figure.

Pew Research Center, 22 July 2025. Researchers tracked the actual browsing of 900 US adults across 68,879 Google searches during March 2025. On pages with an AI summary, 8% of visits produced a click on a normal result. On pages without one, 15% did. Only 1% of visits produced a click on a link inside the summary itself. The full methodology is published.

Chartbeat, across more than 2,500 publisher sites. Pageviews from Google Search fell 34% between December 2024 and December 2025, and 38% in the United States. Chartbeat's own breakdown splits it by publisher size: small sites down 60%, mid-sized down 47%, large down 22%.

Google's position is different, and worth reading in full precisely because it disagrees. In a post published on 6 August 2025 the company said total organic click volume to websites has been "relatively stable year over year" and that click quality has improved. It gives no methodology, no sample, no numbers, and it is now a year old. Treat it as a statement of position rather than as evidence.

What the 58% does not mean

This is the part almost every summary skips, and it changes what you should do.

The 58% is not measured against 2023. It is measured against a forecast of where those same keywords would have landed in December 2025 without AI Overviews. In the same study, comparable informational keywords that show no AI Overview fell from 0.076 to 0.039 over the identical two years.

So position-one click-through rate roughly halved across the board, AI Overviews or not, and then AI Overviews roughly halved it again on the queries where they appear. Featured snippets, local packs and top stories had been doing the same job for a decade. AI Overviews are the sharpest instance of an old pattern, not a new one.

Why this matters practically: if you assume the entire decline arrived with AI Overviews, you will conclude that the fix is to defeat AI Overviews. Half of it was the results page getting denser, and no amount of GEO work reverses that half.

Why the loss is not evenly spread

Two axes decide your actual exposure, and neither is visible in a headline.

By site size. The Chartbeat split is severe: 60% against 22% is not a difference of degree. Large publishers hold more branded demand and more direct traffic, so a smaller share of their total sits in the queries an AI Overview can finish.

By query type. An AI Overview competes for the query it can answer completely on the page. "What is answer engine optimization" is answerable in a paragraph. "Pricing for a GEO agency in Madrid" is not, and neither is your company name. Informational queries absorb the loss. Branded and transactional queries hold.

There is a third finding worth carrying, because it complicates the panic honestly. In Chartbeat's data, total traffic to those publishers stayed roughly steady while search fell. Internal navigation went from about 38% of pageviews in 2024 to about 41%, and dark social from about 7% to about 10%. The audience did not disappear. Its route changed.

How to tell which of your pages is exposed

You cannot read this directly, and knowing why saves you a week of looking for a report that does not exist.

Google folds AI Overviews and AI Mode into your ordinary Web search numbers. Google's own documentation states that sites appearing in AI features are included in overall search traffic in Search Console. There is no filter that isolates them in the main Performance report.

Since June 2026 there is a dedicated Generative AI performance report, covering AI Overviews and AI Mode. Read its limits before you plan around it. It reports impressions only, with no clicks and no position. It excludes Search Labs experiments. It is rolling out to a subset of sites, so you may not have it at all.

That leaves the shape of your own data, which is enough. In Search Console:

1. Open the Performance report and compare the last 6 months against the same period a year earlier.
2. Add the Pages dimension and sort by impressions.
3. Look for the specific combination: impressions flat or up, average position flat, clicks and CTR down.

That pattern means you are still being shown and no longer being clicked, which is what displacement looks like from your side. A page losing impressions and position is a different problem, usually ranking, and the fix is different too.

Sort what you find by intent. Informational pages showing that pattern are doing their job in a world where the job pays less. Commercial pages showing it are the ones to act on first.

What actually works in response

The goal stops being the click and becomes the mention, because the query is answered on the results page whether or not you are one of the sources named.

Be the cited source rather than the displaced one. An AI Overview lifts passages. Write blocks that stand alone, answer one question completely, and survive being quoted with no surrounding context. Our guide to getting cited by ChatGPT covers the passage structure in detail.

Move effort toward queries an overview cannot finish. Comparisons with real trade-offs, pricing with conditions, anything specific to your market or your data. If a paragraph can end the search, it will.

Report on citations, not only on rank. If you measure this quarter the way you measured 2023, you will record a loss and miss the offsetting gain. You can check where you currently stand with our free AI visibility checker.

Keep the branded surface healthy. It is the part of your traffic that holds, and it is the part that grows when an assistant names you and the person then searches your company.

What does not work

Blocking AI crawlers as a defensive move. Worth updating your assumptions here, because the situation changed in 2026. Google now offers a Search generative AI control in Search Console under Settings, and states that it "isn't used as a ranking or inclusion signal affecting other parts of Search". So the old objection, that opting out costs you your rankings, no longer applies to Google. The real objection is simpler: the overview appears either way, so you give up the citation and get nothing back. Note also that this control does not stop training, which is a separate directive.

Chasing the 58% as if it were reversible. Half of it predates AI Overviews. Setting a target of returning to your 2023 click-through rate sets you up to fail at something no one has achieved.

Rewriting pages that were never exposed. If your commercial pages held their clicks, the decline you read about did not happen to you, and rewriting them to defend against it is work with no upside.

Treating Google's "stable click volume" statement as reassurance. It may be true in aggregate across the whole web and still be false for your site. An aggregate that holds while small publishers lose 60% is arithmetic, not comfort.

Where this goes next

The measurement gap is the thing to watch. Google now reports AI impressions but not AI clicks, which means the industry can see exposure and not outcome. Whether that report ships clicks, and whether it reaches every site, decides how much of this remains inference.

Every figure in this post was checked against its primary source on 12 August 2026. If any of them is later corrected at the source, we will correct it here and say what changed.