Google AI Overviews can reduce clicks to conventional search results for searches where an overview appears, but the available evidence does not establish one traffic-loss rate for every website. User observations, Google’s aggregate reporting and a field experiment measure different things. For a publisher, the practical question is whether clicks, visits and valuable reader actions are changing on the pages and queries most exposed to overviews—not whether a single headline percentage applies to the whole site.
What the evidence says about clicks
The clearest figures describe distinct populations and outcomes. Read them as evidence about specific searches or settings, not as a forecast for an individual publisher.
| Evidence | What was measured | What the result does—and does not—show |
|---|---|---|
| Pew Research Center, July 2025 | In Pew’s analysis of Google searches in March 2025, 58% of respondents had at least one search that produced an AI-generated summary. Users clicked a traditional result on 8% of visits with a summary, compared with 15% of visits without one. They clicked a link within a summary on 1% of visits to pages with one. | These are observed user behaviors, not a guaranteed causal change in any site’s traffic. The 8% and 15% figures concern clicks on traditional results; the 1% figure concerns summary links. |
| Google, August 6, 2025 | Google said overall organic click volume from Search to websites was “relatively stable year-over-year” and that it was sending slightly more “quality clicks,” defined as clicks after which users do not quickly click back. | This is Google’s account of aggregate Search traffic and its own quality measure. It is not an independent estimate of AI Overview effects, nor a report about any one publisher. |
| Saharsh Agarwal and Ananya Sen, field-experiment paper posted April 3, 2026 | In the study setting, conditional on an AI Overview appearing, the authors report that outbound organic clicks fell 39.8% and zero-click searches rose 34.5%. | The estimates belong to that experiment’s participants and setting. They should not be presented as a typical publisher-wide loss or extrapolated to all queries, sites or time periods. |
The measures are not interchangeable: a click on a standard result, a click on a cited overview link, an overall organic click, a visit in analytics and a valuable referral are different outcomes. None of the figures above directly states how a particular publisher’s sessions or revenue changed.
Why the findings can differ
The results are not necessarily contradictory. Pew observed what users did on visits with and without a summary. Google reported an all-up year-over-year total across Search, along with its own definition of a quality click. The experiment estimates a conditional effect in its study setting. A change on overview-bearing searches can coexist with stable overall click volume if the population, time window, query mix or outcome differs.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors#1 Best Overall
- Outcome: Impressions, conventional-result clicks, overview-link clicks, site sessions, engagement and revenue answer separate questions.
- Population and geography: A panel of users, selected queries, an experiment and all Google Search traffic are not the same population.
- Product and period: AI Overviews and their link presentation can vary by query, country, language and device, and can change over time.
- Conditioning: A result measured only when an overview appears cannot be directly compared with an all-search aggregate that also includes searches without overviews.
- Method: Observational behavior, company-reported aggregates and experiments have different strengths and limits; an experiment may estimate causality for its participants without establishing a universal long-term effect.
Google has also described an earlier link-placement test: in an August 15, 2024 announcement, it said links shown within AI Overviews were driving higher traffic to publisher sites, but it did not publish an effect size there. That early result is evidence about a particular experiment, not a universal estimate of today’s referrals.
What a referral click does—and does not—tell a publisher
A click count alone cannot establish whether an overview referral is more or less valuable than a traditional organic visit. The UK Competition and Markets Authority’s publisher conduct document records publishers’ concern that they cannot accurately assess the relative value of generative-search referrals. It documents a measurement concern; it does not quantify a difference in referral quality.
Rank #2
To judge business impact, keep the funnel distinct: search impressions show visibility, clicks show outbound activity, analytics sessions show visits that reached the site, and engagement or conversion measures show what those visitors did afterward. A decline at one stage does not by itself establish an equivalent decline at every later stage.
How to investigate a change in your own Google referrals
Use a segmented time series as a diagnostic, not as proof that AI Overviews caused a change. Google Search Console can show search impressions, clicks and click-through rate; site analytics can show sessions, landing pages and downstream engagement. Compare the pages and queries plausibly exposed to overviews with an appropriate baseline or less-exposed group, while preserving the same date ranges and segmentation where possible.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →- Choose the outcome first. Decide whether you are investigating impressions, conventional organic clicks, click-through rate, sessions, engagement or conversions. Do not treat them as synonyms.
- Segment the site. Isolate relevant landing pages and, where available, query groups with substantial informational-search exposure rather than relying only on a sitewide total.
- Compare like with like. Retain a consistent comparison period and review the same pages, query categories and date ranges. Look for divergence between impressions and clicks, and between Search Console clicks and analytics sessions.
- Annotate other changes. Record ranking and query-mix changes, seasonality, site or content updates, and other known shifts that could affect the same metrics.
- Check what visitors do next. Compare engagement or conversion for relevant landing pages and traffic segments if those outcomes matter to your business; click volume alone cannot establish referral value.
A pattern in your own data can identify where to investigate, but it cannot by itself isolate AI Overviews as the cause. The cited evidence does not provide one universally valid attribution procedure for every publisher.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can a publisher opt out of Google’s generative AI Search features?
In a June 3, 2026 announcement, Google said website owners would be able to use a Search Console toggle to control whether a site may appear in and help ground its generative AI Search features. Google said opting out means receiving no traffic or impressions from those features. This concerns those features; the announcement does not say that opting out removes a site from all conventional Google Search results. The control and product behavior can change, so consult Google’s website-owner announcement for current instructions and consequences before changing a site setting.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
