> ## Content Index
> Fetch the complete content index at: https://varops.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Google’s AI Overviews cut outside clicks 40% on searches that show one and 18.5% across all searches. Here’s what you can do about it.
- URL: https://varops.com/googles-ai-overviews-cut-outside-clicks-40-on-searches-that-show-one-and-18-5-across-all-searches-heres-what-you-can-do-about-it/
- Published: 2026-11-18T12:07:11.000Z
- Updated: 2026-11-18T12:07:11.000Z
- Description: Google says the clicks it sends websites are stable and better. A randomized trial disagrees on quality, and the 40% its critics quote is a per-search number, not a per-site one.
- Author: Nix Nullty
- Tags: Overhyped

*Yesterday Gritt put two runs of the same eval side by side and found the answer had changed with the model. Today the two numbers come from Google and from a randomized trial. Google says the clicks it sends to websites have held steady and got better. Two researchers hid AI Overviews from a random group of volunteers and counted.* [***Nix Nullty***](https://varops.com/columnist/nix/) *tests Google's claim first, then turns the same test on the 40% figure its critics now quote. The piece ends with a sum any team can run on its own Search Console data this week. —* [*Muximus*](https://varops.com/columnist/muximus/)

---

Google's head of Search says the clicks it sends to websites have been "relatively stable" year over year as AI Overviews spread, and that the clicks it still sends are better ones. A preregistered randomized trial hid the Overviews from a randomly chosen group of volunteers and found people clicked through to outside websites about 40% less when one appeared, and that the clicks they did make were no better.

Verdict: Google's better-clicks story fails the test. The 40% that critics now wave back at Google is real, but it is a per-search number, and people keep spending it as a per-site one. RevScope's November 9 roundup ran it as "Google's AI Overviews cut clicks to websites by 40%." Any single site's number is the per-search effect times how much of its traffic comes from searches that show an Overview. That one can be worked out in Search Console.

## The claim, at full strength

On August 6, 2025, Liz Reid, VP and Head of Google Search, wrote that "total organic click volume from Google Search to websites has been relatively stable year-over-year." Google was "actually sending slightly more quality clicks to websites than a year ago," she added, and a quality click is one where the user doesn't quickly click back.

The story hangs together. With Overviews, Google says, people search more, ask longer questions and see more links, so there are more chances to be clicked. Some quick-answer queries end on the results page, as they did with sports scores. Traffic shifts between sites, so some lose and some gain while the total holds. Third-party reports of steep declines, Reid wrote, are "often based on flawed methodologies, isolated examples, or traffic changes that occurred prior to the roll out of AI features."

Credit Where It's Due: that last jab was fair. Most early evidence compared queries that drew an Overview with queries that didn't, or traffic before and after a rollout. Neither design can separate the Overview from everything else that moved at the same time. What the post left out was any number at all - for volume or for quality.

## Somebody ran the experiment

Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon ran the study the critics' data couldn't. It was preregistered with the AEA RCT Registry and posted to SSRN in April 2026 as a working paper. We read the June 17 draft; SSRN lists a July revision we could not open. They recruited 1,065 US-based participants on Prolific - people whose main browser was Chrome on desktop and whose main search engine was Google - and had each install a browser extension. Between January 7 and February 10, 2026, the extension randomly put each person in one of two worlds: Google as usual, or Google with AI Overviews deleted in real time and the results below shifted up to fill the gap. A third group was pushed to Google's AI Mode. So many of them dropped out that the authors treat that arm as exploratory.

Across 396 people in the normal group and 374 in the no-Overview group, the researchers logged 68,089 searches. Overviews triggered on about 41% of them.

When an Overview was triggered, people who saw it made 0.37 clicks to outside websites per search. People who didn't see it made 0.62\. The authors report that as a 39.8% reduction. Searches ending with no outside click went from 54% to 73%. On searches with no Overview, the two groups behaved alike - the check that the extension itself wasn't changing anything.

Then the two results that go straight at Google's argument. People who saw Overviews did not search more; searches per day didn't differ between the groups. And the clicks the no-Overview group made were not junk. Back-to-search rates, bounces (sessions under ten seconds with no interaction) and time on site showed no statistically significant difference. The authors' own words: this "stands in contrast to public claims made by Google about bounce rates coming from AIOs." An exit survey found no measurable difference in satisfaction, perceived quality of information, or ease of finding it either.

The limits are real and they belong next to the number. It is a working paper, not yet peer reviewed. It is desktop Chrome only, people recruited on a paid survey platform, two weeks in early 2026\. Sponsored clicks were too rare - about 0.02 per search - to measure well.

So, the scorecard. The quality claim fails on close equivalents. Google's own definition was clicks "where users don't quickly click back." The trial measured back-to-search for clicks opened in the same tab, plus a ten-second bounce adapted from Google Analytics, and found nothing. That takes out Google's stated reason for the whole shift - that a click after an Overview is worth more. The volume claim is slipperier, because the trial compares two groups at the same moment, not year-over-year totals across the web. But the engine Google named for stable totals, people searching more because of Overviews, didn't show up. Google could end the argument with its own logs. It hasn't published them.

And the trial isn't alone. A second preregistered experiment, by Stephanie Wang, Jeffrey Gleason and colleagues with 1,100 people, run in March 2026 and posted to arXiv in August, also found that removing AI Overviews and AI Mode raised clicks to publishers. We read only its abstract, so it counts here for direction, not for a figure.

**Hype-o-Meter:** Google's stable-and-better-clicks line - 8 out of 10\. [No numbers offered](https://varops.com/that-2x-developer-productivity-number-who-owns-it-and-why-nobody-can-verify-it/), and the one part a randomized trial could test came back flat.

## Now do the critics

Here's where the 40% crowd gets the same treatment. The 39.8% applies only to searches where an Overview shows up. The paper converts it in a footnote: at the 41% trigger rate in its sample, the average cut in outside clicks across all searches is about 18.5%. Less than half the headline, from the same study.

It isn't spread evenly, either. Informational queries were 71% of searches and drew an Overview 53% of the time. Navigational queries - people typing a brand or a site name - were 18% of searches and drew one 15% of the time; transactional queries, 12% and 6%. The effect was large and significant for informational queries. For navigational and transactional ones it couldn't be told apart from zero, partly because those samples were small.

A site that answers questions and a site people search for by name are standing in different weather. All About Berlin, a guide site Nicolas Bouliane has run since 2017, is the first kind. Bouliane says it has lost 75% of its traffic to AI Overviews. That is his account of his own analytics, not a causal measurement. It is also exactly the shape the trial predicts for an informational site.

**Hype-o-Meter:** Reading the 40% as a site's loss - 6 out of 10\. The number is real. It just belongs to a search, not to a site.

## The 41% is already out of date

The trial's trigger rate is an early-2026 number. DataForSEO, which sells search data, reports Overviews on 61.43% of the Google results it analyzed, up from 5.93% two years earlier; RevScope dates those endpoints to August 2026 and September 2024\. By RevScope's reading of the same report - RevScope sells revenue analytics, and the full report sits behind a sign-up form we didn't fill in - informational keywords went from 48% in late 2025 to 70% in the third quarter of 2026, and transactional from 11% to 29%.

The authors saw this coming. The aggregate impact "depends critically" on how often Overviews appear, they wrote, and could become "substantially larger" even if the per-search effect holds still. Whether it has held is something we could not verify: we found no randomized estimate from after March.

## The version any site can run

Google won't hand this number over. Search Console's generative AI reports, announced in June for an initial subset of sites, show impressions, pages, countries and devices for AI features. Clicks aren't on the list.

So build it:

1. Export the top queries by clicks from Search Console for the last three months.
2. Run each one in a logged-out US search and mark whether it shows an AI Overview.
3. Add up the share of clicks that come from queries with an Overview. Weight by clicks, not by query count - clicks are rarely spread evenly across a site's queries.
4. Multiply that share by about 0.4\. A site that gets 80% of its Google clicks from Overview queries is looking at roughly a third fewer clicks than it would get without Overviews; a site at 20%, around a tenth. Today's clicks already carry the cut, so this understates the loss, most of all at low shares. This is our arithmetic, and it assumes the trial's effect fits the site's query mix - least safe for navigational and transactional queries, where the trial found no clear effect.
5. Treat the answer as a desktop estimate. The trial was desktop only, and there is no randomized number for phones yet.

We get search traffic too, so this sum is not an abstraction for us either.

When an agency or vendor pitches a recovery plan on the strength of a flat 40% loss, ask two things back: the share of your clicks that come from queries showing an Overview, and the effect they assumed for people who search for you by name. If they can't answer the first, they are selling against a headline, not against your traffic.

Plan with the site's own exposure times the per-search effect, and keep score in revenue per visit, not rank or visit counts. A page can hold position one and still earn fewer clicks than it did a year ago.

From thinking to doing. 

[![CTA Image](https://cdn.varops.com/content/images/2026/10/logo-black.jpeg)](https://varops.com/transform/) 

****Less hype. More results.**

AI promises are everywhere. We prefer results. VarOps helps companies cut through the hype, avoid expensive detours, and build AI systems that deliver measurable business value.

[Get Beyond the Hype → ](https://varops.com/transform/) 

## Sources

- Saharsh Agarwal and Ananya Sen, "The Impact of Google AI Overviews on Publisher Traffic and User Experience: Evidence from a Field Experiment" (SSRN working paper; June 17, 2026 draft): https://papers.ssrn.com/sol3/papers.cfm?abstract\_id=6513059
- Liz Reid, Google, "AI in Search is driving more queries and higher quality clicks" (Aug 6, 2025): https://blog.google/products/search/ai-search-driving-more-queries-higher-quality-clicks/
- Stephanie T. Wang, Jeffrey Gleason and colleagues, "AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence" (arXiv, Aug 2026): https://arxiv.org/abs/2608.18352
- DataForSEO, Search Insider 2026 (Oct 2026): https://dataforseo.com/whitepapers/search-insider-2026
- RevScope, "Google's AI Overviews cut clicks to websites by 40%" (Nov 9, 2026): https://revscope.co/blog/search-seo-geo-2026/
- Nicolas Bouliane, "AI is killing All About Berlin": https://nicolasbouliane.com/blog/death-by-ai
- Google Search Central, "Introducing Search Generative AI performance reports in Search Console" (June 3, 2026): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports