AI Overviews Statistics 2026: Search Impact Data

Google’s AI Overviews now reach over 2.5 billion monthly active users, which the Washington University measurement team calls arguably the most widely encountered deployment of generative AI. Independent measurements of what those users then do arrived through 2026, and they disagree sharply on one number while agreeing closely on another.

The share of searches that trigger an AI Overview ranges from 13.7% to 36% depending on who is counting and which queries they count. The click effect is far more consistent: Pew Research Center measured a traditional-result click on 8% of visits to search pages carrying an AI summary, while visitors without one clicked a result nearly twice as often, on 15% of visits.

Key Takeaways

• Google reports AI Overviews at over 2.5 billion monthly active users, alongside 13 Google products that each pass a billion users.
• The activation-rate spread between studies is a sampling artifact: a crawler issuing 55,393 trending queries recorded 13.7% while a panel watching real searches recorded 36%.
• Question phrasing is the single strongest activation predictor, at 64.7% for question-form queries versus 9.5% for non-question queries.
• AI Overviews cite sources that search itself does not rank: nearly 30% of cited domains do not appear in the co-displayed first-page results at all.
• Accuracy is not uniform across the answer: 11.0% of 98,020 atomic claims were unsupported by the pages the summary cited.
• The feature is not universally extractive. Reddit communities marked Safe-for-Work (SFW) saw daily comments rise by 12.0% relative to communities excluded from AI Overviews.

Editor’s Choice

• AI Overviews monthly active users: over 2.5 billion.
• AI Mode monthly active users: surpassed 1 billion, one year after debut.
• Overall activation rate, trending-query crawl: 13.7% across 55,393 queries.
• Traditional-result click rate with an AI summary: 8% of all visits.
• Click rate on links inside the summary: 1% of visits.
• Unsupported atomic claims: 11.0% of 98,020.
• Highest category activation rate: 46.1% in Hobbies and Leisure.

AI Overviews Reach and Monthly Active Users

• Google put AI Overviews at over 2.5 billion monthly active users in its May 19, 2026 I/O keynote post.
• AI Mode, the conversational surface, surpassed 1 billion monthly users just one year after its debut.
• Google counts 13 products with over a billion users each, five of them above 3 billion.
• The Washington University measurement team describes AI Overviews as reaching over 2 billion users who may not realize the answers they see are AI-generated.
• Google reported that Search queries reached an all-time high in the quarter before I/O 2026, which complicates the simple substitution story.
• AI Mode queries have more than doubled every quarter since launch.

Passive deployment is what separates AI Overviews from a chatbot. Nobody opts into an AI Overview; it arrives on the results page whether the searcher wanted a synthesized answer or a list of links, which is why reach figures here are not comparable to a product people choose to open.

Scale of this kind turns small percentage effects on clicks into large absolute traffic movements. That pattern is visible in our Google usage data and in the per-surface breakdown behind Google AI Mode.

Source: Google I/O 2026 keynote post, May 2026

How Often AI Overviews Appear in Google Search

• The Washington University crawl found 7,583 of 55,393 trending queries returned an AI Overview, an activation rate of 13.7%.
• The Penn and Northeastern field experiment observed 36% of participant searches triggering AI Overviews across real everyday browsing.
• AI Mode barely registered in the same panel: participants used AI Mode for 0.6% of their searches.
• Pew Research Center found that for 58% of its 900-person panel, at least one search in March 2025 produced an AI-generated summary.
• Within that panel, 12,593 of 68,879 unique Google searches produced an AI summary.

The two headline activation numbers are not in conflict, and treating them as rival estimates of one quantity is the most common error in coverage of this topic. A crawler issuing trending queries samples the head of the distribution, where sports scores and entertainment queries dominate. Browser panels instead capture whatever real people typed, which skews toward question-shaped informational queries.

Any honest answer to the activation question is a range with the query mix attached. Quoting a single figure quietly chooses a query mix on the reader’s behalf.

Don’t chase tech news. We track it for you.

One weekly briefing with the launches, AI developments, and breaches that matter. No filler.

Email address

Subscribe Now

Read by pros at Fortinet, TSMC, Barclays, and Deloitte.

What percentage of Google searches have AI Overviews?

Measured activation runs from 13.7% across 55,393 trending queries to 36% of searches in a 1,100-person browsing panel. The gap is a query-mix difference, not a disagreement: trending queries skew toward sports and entertainment, while real users type more question-shaped informational queries, and those trigger overviews far more often.

Recent Developments

• April 2026: Over a 40-day window ending April 21, 2026, a crawler issued 55,393 trending queries across 19 topical categories, of which 7,583 returned an AI Overview.
• March 2026: Between March 17 and March 19, 2026, researchers recruited US-based Chrome users in waves for a preregistered field experiment of 1,100 participants on Google Search.
• May 2026: On May 19, 2026, Google disclosed AI Overviews to over 2.5 billion monthly active users at I/O 2026.
• May 2026: On May 20, 2026, Google merged AI Overviews and AI Mode into one AI Search experience, live across desktop and mobile worldwide.
• June 2026: On June 20, 2026, the Reddit engagement study reached version 3, reporting a 12.0% rise in daily comments in Safe-for-Work communities.
• September 2026: On September 2, 2026, the Wikipedia impact paper reached version 6, reporting declines of 5.45% and 4.82%.

Google also expanded the personalization layer, taking Personal Intelligence in AI Mode to nearly 200 countries and territories across 98 languages with no subscription required. The same consolidation wave is visible across agentic products tracked in our AI agents data.

AI Overview Activation by Topic Category

• Activation varies by a factor of 13 across categories, from 3.5% in Beauty and Fashion to 46.1% in Hobbies and Leisure.
• Sports supplied the largest query volume in the crawl, 28,446 queries at a 9.8% activation rate.
• Science ran unusually hot at 39.9% activation across 1,560 queries.
• Politically sensitive categories fall below average: Politics at 7.5% and Law and Government at 9.6%.
• Health broke the sensitive-topic pattern at 26.6% activation, which the researchers note shows suppression is not applied uniformly.

Category spread matters more than the average for anyone estimating exposure on their own topics, because a publisher sitting in Hobbies and Leisure faces roughly thirteen times the activation rate of one sitting in Beauty and Fashion. Health is the figure worth arguing about: it sits well above the political and legal categories where Google applies visible caution, yet health queries carry comparable consequences for the person asking.

Question-Form Queries and AI Overview Triggering

• Question phrasing is the dominant trigger: 64.7% of question-form queries returned an AI Overview versus 9.5% of non-question queries, a 6.8-fold difference.
• The gap is driven by phrasing: question-form queries returned an AI Overview 64.7% of the time against 9.5% for non-question queries.
• The overall activation figure of 13.7% conceals that 64.7% of question-form rate, which is why aggregate percentages mislead.
• Politically sensitive topics see markedly lower rates even when phrased as questions.

Worth noting: The question-form gap does most of the work in explaining why activation estimates diverge. Question-form queries return an AI Overview 64.7% of the time against 9.5% for other phrasings, so a crawl of trending queries catches few interrogatives while a real-user panel catches many. The query mix moved, not the algorithm.

Click-Through Rates With and Without AI Overviews

• Pew Research Center measured a traditional-result click on 8% of visits to search pages carrying an AI summary; visitors to pages without one clicked a search result nearly twice as often, on 15% of visits.
• Links inside the summary itself were followed on just 1% of all visits to pages with such a summary.
• The randomized experiment found assignment to AI Mode Search reduced click-through rate by 18.8 percentage points, with a 95% confidence interval from 22.2 to 15.3 percentage points.
• Running the experiment in the other direction, removing AI Overviews and AI Mode increases click-through rates to publishers.
• Compliance in that experiment was partial: 51.1% of AI Overviews were successfully hidden, and the median participant had 50% hidden.

Key finding: Pew Research Center’s browsing panel of 900 US adults recorded a traditional-result click on 8% of AI-summary visits versus 15% without, and just 1% on the summary’s own citations. The summary absorbs the click rather than redistributing it, which is the mechanism publishers describe.

Do people click on links inside AI Overviews?

Rarely. Pew Research Center recorded a click on an in-summary link on 1% of all visits to search pages carrying an AI summary, against 8% for traditional result links on the same pages. The citations are visible and technically clickable, but they function as provenance markers rather than as a traffic channel, which is why AI Overview citation counts track poorly against referral logs.

What Users Do After Seeing an AI Overview

• Browsing sessions ended on 26% of pages with an AI summary against 16% of pages with only traditional results.
• The behavioral baseline is broad AI exposure, not AI seeking: 93% of panellists visited at least one page mentioning AI, while around 49% visited a page primarily focused on AI.
• Deliberate AI research was rarer still, at one-in-ten panellists running an AI-related query on a search engine during the month.

Session-end rates are the most under-quoted numbers in the whole dataset, because it separates two very different failure modes for a publisher. A searcher who clicks a competitor still stays in the open web; a searcher who closes the browser has left it. The gap between the two session-end rates is 10 percentage points, and that loss is invisible in any referral report. Device mix shapes this too, since mobile sessions end faster than desktop ones across the Android install base.

AI Overview Impact on Publisher and Reference Traffic

• Default AI Overview availability reduced English Wikipedia search traffic by 5.45% against the German benchmark and 4.82% against the French one.
• That estimate is version-dependent. The paper reached version 6 on September 2, 2026, after a first version on February 5, 2026 and a withdrawn second version.
• The identification strategy compares monthly external-search referrals to English Wikipedia articles with referrals to the same articles in German and French, exploiting the staggered geographic rollout.
• The randomized experiment measured the direction independently, finding that removing AI Overviews and AI Mode increases click-through rates to publishers.
• The same experiment found AI Mode specifically reduces click-through rates and erodes user experience and trust in information found on Google.

By the numbers: The Wikipedia difference-in-differences design put the referral loss at 5.45% against German articles and 4.82% against French ones. Both figures come from the paper’s current version 6, dated September 2, 2026, rather than its first draft in February.

Anyone citing preprint effect sizes in this area should check the version number before the number. This Wikipedia paper alone moved through several numbered versions, including a withdrawn one, before reaching its current figures. That version history is worth carrying into any SEO forecast built on these papers.

Where AI Overviews Get Their Sources

• Nearly 30% of AI Overview-cited domains do not appear in the co-displayed first-page results at all, which points to a selection mechanism separate from ranking.
• AI Overview-cited domains are more credible than the co-displayed first-page results on the study’s credibility measure.
• Wikipedia, YouTube and Reddit are the most frequently cited sources in both Google AI summaries and standard search results.

A citation pool that is 30% disjoint from page one is the single most actionable finding here for publishers, because it means ranking and being cited are partially independent goals. The heavy presence of user-generated platforms alongside YouTube also explains why so many AI answers read as community consensus rather than as institutional reporting.

Source: arXiv 2605.14021 and Pew Research Center, 2026

AI Overview Accuracy and Unsupported Claims

• Decomposing answers into atomic assertions, 11.0% of 98,020 claims were unsupported by the pages the summary cited.
• Omission is the dominant failure mode, meaning the cited page simply does not address the assertion rather than contradicting it.
• Source quality and claim fidelity are largely independent, so a well-sourced overview is not reliably a faithful one.
• The claim set was drawn from the same 55,393-query, 40-day crawl that produced the 13.7% activation figure.

The takeaway: Across 98,020 atomic claims extracted from AI Overviews, 11.0% were unsupported by the cited pages, and the researchers report that source quality and claim fidelity move independently. Better citations do not imply a more faithful answer, which breaks the intuition most quality heuristics rest on.

Independence between the two is the finding with the longest tail. It means a reader cannot audit an AI Overview by inspecting its sources, because the answer’s fidelity to those sources is a separate variable. That failure shape differs from the confident-invention pattern documented in our LLM hallucination statistics and in the head-to-head AI model comparison data, where the model asserts something no source supports at all.

The Content Types AI Overviews Help Rather Than Hurt

• Safe-for-Work Reddit communities recorded daily comments up 12.0% and commenting users up 12.4% relative to Not-Safe-for-Work communities.
• The identification rests on Google’s own moderation rule: Safe-for-Work communities are surfaced in AI Overviews while Not-Safe-for-Work communities are prohibited from being referenced in AI Overview summaries, though both are indexed by organic search.
• The gains were concentrated in experience-based discussions such as opinions, advice, and personal experiences rather than fact-based information.
• The effect did not persist: the subsequent introduction of Google AI Mode largely eliminated these gains in experience-based content.

This is the result that breaks the universal-decline narrative, and it does so with an unusually clean natural experiment. When a summary can restate a fact, it substitutes for the source. Where the value is a human recounting an experience, the summary advertises the source instead. Reversal under AI Mode suggests the boundary is an interface property rather than a content one, which is a fragile thing to build on.

Similar substitution-versus-referral splits show up in our AI in social media data alongside assistant tools such as ChatGPT.

Source: arXiv 2605.16428, 2026

What AI Overviews Cost Publishers in Ad Revenue

• Well over half of AI Overview-cited pages carry display advertising, so a suppressed click is a measurable revenue loss rather than a notional one.
• The asymmetry is explicit in the research: Google’s own sponsored ads continue to appear on the same page while organic clicks fall.
• Publishers lose revenue when AI Overviews suppress the click-through, which is the mechanism the researchers name.
• Scale converts that into real money. At over 2.5 billion monthly active users, a single-digit click shift moves a very large absolute number of sessions.

Advertising asymmetry is what turns an interface change into an economic transfer. Both parties monetize attention on the same results page, and only one of them needs the click. Publishers weighing headcount against that shift may find our AI job loss statistics a useful companion read.

Source: arXiv 2605.14021 and Google I/O 2026 keynote post

How accurate are AI Overviews?

Accuracy is high in aggregate and unevenly distributed. A large-scale audit decomposed AI Overview responses into 98,020 atomic claims and found 11.0% unsupported by the cited pages, with omission the dominant failure mode. Omission is a gentler error than contradiction: the summary asserts something the cited page does not address, rather than something it refutes.

The harder finding for readers is that source quality and claim fidelity are largely independent. A reader cannot infer that an overview is faithful from the fact that it cites reputable pages, because the two properties do not track each other. On a health or finance query, that argues for opening the citation rather than trusting the synthesis, and Health queries return an AI Overview 26.6% of the time, well above the political and legal categories where Google applies visible caution.

How can publishers measure the impact of AI Overviews on their own traffic?

No analytics product reports AI Overview impact directly, so every credible estimate published so far rests on a comparison design. The Wikipedia study compared English articles against the same articles in German and French, finding declines of 5.45% and 4.82%. The field experiment randomized the interface itself and found that removing AI Overviews and AI Mode increases click-through rates to publishers.

Three practical implications follow for a publisher’s own measurement. Segment by query shape first, since question-form queries trigger AI Overviews at 64.7% versus 9.5% for non-question queries, which means impression loss concentrates in question-shaped keyword sets. Track session abandonment rather than referrals alone, because 26% of AI-summary pages ended the browsing session against 16% without one, and that loss never appears in a referral log.

Benchmarking comes last. Compare against assistant surfaces such as Perplexity, and read any single-percentage activation claim against the wider Google search statistics picture before building a forecast on it.

Conclusion

AI Overviews reach over 2.5 billion monthly active users, and the 2026 research record now describes what happens on that surface with reasonable precision: activation somewhere between 13.7% and 36% of searches depending on query mix, a traditional-result click rate of 8% of visits where a summary appears against 15% where visitors clicked a result nearly twice as often, and 11.0% of 98,020 audited claims unsupported by their own citations. Direction is consistent across every independent design, though magnitudes are still moving between preprint versions.

Two of these numbers are actionable for publishers, and one is not. Activation and click loss vary by query shape, which makes them addressable through content and keyword strategy. Claim fidelity is Google’s variable, not anyone else’s, and the finding that it moves independently of source quality means no amount of publishing rigor upstream guarantees a faithful restatement downstream.