Google AI Overviews: Why Website Clicks Are Falling14 min read2,711 words

Google AI Overviews: Why Website Clicks Are Falling

Google AI Overviews can reduce outbound clicks by about 40%. See what the evidence says and how to adapt SEO, GEO and business measurement.

Google AI OverviewsAI search trafficgenerative engine optimizationSEO strategyAI citations
Google AI Overviews: Why Website Clicks Are Falling

Google AI Overviews Are Changing the Economics of Search

For years, a high Google ranking was a decent stand-in for visibility. It wasn't a promise of traffic - nothing is - but the pattern was simple: show up near the top, earn clicks, and convert some visitors. AI Overviews poke a hole in that assumption. Google now answers a lot of searches right on the results page, so users can get what they came for without visiting a website.

So the real question isn't whether SEO is "over." It's whether the clicks and revenue behind your SEO work are shifting - and what you can do in response. A randomized desktop experiment found that searches showing an AI Overview produced about 40% fewer clicks to outside websites than similar searches where the Overview was hidden. At the same time, AI citations don't yet appear to fully replace those lost visits at scale.

This article sticks to the strongest evidence and leaves the louder industry claims to the side. We'll look at how widely AI Overviews show up, what the click research can (and can't) prove, what generative engine optimization (GEO) adds to SEO, and how to make decisions based on business outcomes - not rankings alone.

How widely are Google AI Overviews appearing?

An AI Overview is a generated answer placed above or among Google's regular results. It can summarize information from multiple pages and include links to some of its sources. For businesses, that creates two routes to outcomes: the searcher clicks through to one of the cited pages, or they get enough from the summary and move on.

DataForSEO's panel of tracked results shows how fast the feature is spreading. AI Overviews appeared on 5.93% of tracked results in September 2024, rising to 61.43% in August 2026. That's not limited to a few test queries - it's a substantial change to the search results landscape. Pew Research Center's separate look at US searches lines up as a cross-check: it found summaries on 18% of searches it reviewed in March 2025, close to DataForSEO's numbers around the same time.

No keyword category is reliably sheltered

This isn't confined to purely informational questions like "how does compound interest work?" DataForSEO reported increases across all four broad search intents from the fourth quarter of 2025 to the third quarter of 2026:

Search intent Q4 2025 Q3 2026
Informational 48.42% 70.44%
Commercial 26.30% 51.35%
Navigational 17.04% 34.03%
Transactional 10.82% 28.64%

Informational searches are still the most exposed, but commercial and transactional queries are clearly in the mix. And low volume isn't a safety net either. In the third quarter of 2026, keywords with 201-300 monthly searches showed Overviews more often than keywords with more than 600 searches.

Industries also differ. DataForSEO's report puts health and finance near 70%, while apparel, dining, and real estate are lower. Still, that variation doesn't create a dependable escape route. The wording of the query, the user's intent, the sector, and Google's evolving product choices all matter.

Search features are being rearranged, not just added

As AI Overviews have become more common, some older result types have gotten pushed down. In the same DataForSEO series, featured snippets dropped from 8.96% of results in September 2024 to 0.44% in August 2026. Discussion and forum blocks also declined over that period. Video and local map results moved differently, with their own changes and fluctuations.

The practical takeaway is that a strategy built around one feature - or one kind of keyword - can get fragile. The safer assumption is that the results page changes often enough that old click-through rates are a weak forecast of future traffic.

Do AI Overviews reduce clicks to websites?

The clearest evidence comes from a randomized field experiment by Saharsh Agarwal and Ananya Sen. US desktop Chrome users were randomly assigned to one group that saw Google normally and another group whose extension hid AI Overviews. The researchers then observed real searches over two weeks in January and February 2026.

For searches that would have displayed an Overview, users who could see it averaged 0.37 clicks to external websites per search. Users who couldn't see it averaged 0.62. The authors translate that into a 39.8% reduction in outbound clicks. They also found that searches with no outbound click rose from 54% to 73% when the Overview was visible.

This is stronger than simple click-through comparisons, because random assignment helps separate "the feature changed" from "the kind of search changed." A second preregistered experiment by Stephanie T. Wang, Jeffrey Gleason, and colleagues reached a similar conclusion: removing AI-generated search features increased clicks to publishers.

What the experiment does - and does not - tell you

That "about 40%" number applies to searches where an AI Overview would appear, not every Google query. Across all searches in the Agarwal and Sen analysis, the estimated reduction was about 18.5% of organic clicks. The effect held up statistically for informational searches. For navigational and transactional searches, the sample sizes were too small to draw firm conclusions.

There are also real limitations. The study is a working paper, not a peer-reviewed publication. It focuses on desktop Chrome users in the US, and it covers only two weeks. Still, among the available evidence, it's the strongest causal result in the material - and it shouldn't be treated as a universal forecast for every site, device, or market.

Pew Research Center's browsing data points in the same direction, but it's less clean on cause. People clicked an ordinary search result on 8% of visits to a page with an AI summary, versus 15% without one. They also ended their browsing session on 26% of summary visits, compared with 16% without a summary. Because the searches that trigger summaries can differ from other searches, this comparison can't isolate the summary's effect as confidently as a randomized test.

Rankings and clicks are no longer interchangeable

DataForSEO also reported a drop in click-through rate for the first organic result on desktop - from 17.83% in September 2025 to 7.43% in August 2026. That trend is worth watching, but the data doesn't explain why it happened. For example, the series shows a large desktop drop between February and March 2026 while the reported prevalence of Overviews changed only slightly. Layout changes, shifts in user behavior, or changes in the queries being measured could all play a role.

Google's head of Search, Liz Reid, has said total organic clicks are relatively stable year over year and that clicks are becoming higher quality. Google hasn't published figures in support of that claim. So businesses should treat both broad click estimates and Google's framing as context - not as a replacement for tracking channel and revenue performance in their own analytics.

What does GEO change about SEO?

Generative engine optimization, or GEO, is the practice of improving the odds that your content gets cited or used in AI-generated answers. Based on what's available so far, GEO doesn't look like a totally separate discipline with its own reliable set of technical tricks. It looks a lot like SEO - plus extra attention to what AI systems can find, and how easily your pages' information can be understood and quoted.

Google says pages still need to be indexed and eligible to appear with a snippet, and it says there are no additional requirements specific to AI Overviews. That's Google's description of its own system, not independent proof that every eligible page has an equal chance of being cited. Even so, external research hasn't produced a dependable checklist that overrides the basics.

Authority still appears to matter, but correlation is not causation

DataForSEO found a moderate relationship between the number of referring domains (how many distinct websites link to a site) and the frequency of AI citations. The correlation was 0.406 for AI Overviews and 0.377 for ChatGPT. Links associated with user-generated content had weaker correlations.

This doesn't prove that acquiring links causes AI systems to cite a site. Established publishers typically have more links, stronger brand recognition, and more content - so multiple factors may move together. The analysis also looks at domains that have already been cited, so it can't tell a new or small site exactly what it takes to earn its first citation.

In practical terms, the takeaway is modest but still useful: legitimate editorial links and a credible web presence remain sensible long-term investments. They aren't guaranteed GEO levers, and buying links or chasing volume without regard to quality isn't supported by these findings.

Make pages useful to quote, not merely optimized to rank

Page-level studies point toward clarity and evidence over keyword gymnastics. A SISTRIX study paired cited pages with uncited pages from the same site to help control for differences in domain strength. In the sample, a current year in the title was associated with citations 2.7 times as often, and a table of contents with jump links was associated with citations 1.7 times as often. The authors treat these as correlations and explicitly caution against treating them as guaranteed tactics.

There's also an earlier GEO experiment by Aggarwal and colleagues that tested specific content changes. On its benchmark, adding quotations, statistics, and cited sources increased visibility in generated answers, while keyword stuffing reduced it. Because it used an older model and a setup that doesn't map perfectly to today's search products, the exact percentages shouldn't be used as a forecast. But the direction is more actionable than the numbers: answer the question clearly, support claims with evidence, and make key facts easy to extract accurately.

There's no convincing evidence that an llms.txt file solves this. SE Ranking found no correlation between the file and citations across nearly 300,000 domains; SISTRIX also reported no measurable effect, with methodological limits. Google says AI-specific files or markup aren't required for its search features. If you already maintain one, it likely won't hurt, but it shouldn't displace work on content quality, technical accessibility, and measurement.

How should businesses adapt their search strategy?

Giving up on SEO isn't the right response, just because clicks are under pressure. Search still drives far more visits than AI assistants. Ahrefs estimated that AI assistants accounted for 0.28% of traffic across 74,752 websites in March 2026, compared with roughly 345 million visits from Google and 3.5 million from AI assistants in its dataset. That share may be off if app traffic is classified as direct traffic, and the mix will change as more assistant experiences roll out. But for now, AI referrals don't replace organic search at scale.

A better approach is to protect what still creates value in search while testing how new surfaces affect discovery and decisions.

Use a practical four-part operating plan

  1. Keep pages technically eligible. Make sure important pages can be crawled and indexed, render their core content reliably, and show up in ordinary search results. If Google can't access a page, AI visibility shouldn't be your top priority.
  2. Review citations for important topics. Check how Google AI Overviews and major assistants describe your category, products, and expertise. Note which publishers, communities, and reference sites appear. Treat this like qualitative research, not a stable ranking report; citations and answers can shift.
  3. Improve the source material. Give each page a clear purpose, direct answers, descriptive headings, and evidence a reader can verify. Use original data where you have it, cite credible sources, and keep time-sensitive information current. For editorial content, a meaningful update date can help readers judge freshness.
  4. Measure outcomes by channel and landing page. Track qualified leads, purchases, revenue per visitor, and assisted conversions alongside sessions and rankings. Separate Google organic traffic from identifiable AI referrals where your analytics setup allows it. A smaller channel can still be valuable if it brings unusually relevant visitors - but you have to prove that with your data.

Build a measurement loop around business value

Search Console can show performance for Google search, but Google includes AI feature traffic in its broader web reporting. Google announced generative AI reporting in Search Console in 2026, describing visibility metrics like impressions, pages, countries, and devices. Impressions can tell you that a URL appeared. They don't tell you how many visits or sales that appearance produced.

A monthly review can stay simple. For each important landing page, compare organic impressions, clicks, conversion rate, and revenue with the previous period and the same period last year. Add notes for major product or site changes. Then look at queries and pages where impressions stay steady but clicks fall. That pattern can signal a results-page change, but it isn't proof by itself - seasonality, ranking shifts, and demand changes can look the same.

If you can identify referrals coming from ChatGPT, Claude, Perplexity, or other assistants, evaluate them separately by quality. Ahrefs reported that AI visitors were 0.5% of its site's visitors but 12.1% of signups. That's useful as an example of channel differences - not a universal conversion benchmark. One SEO software company can't represent every business. Use it as a reason to measure carefully, not as justification for assuming exceptional conversion rates everywhere.

Key takeaways

  • A randomized desktop study found about 40% fewer outbound clicks on searches that displayed an AI Overview; across all searches in the study, the estimated loss was about 18.5% of organic clicks.
  • AI Overviews show up across informational, commercial, navigational, and transactional queries, though exposure varies by industry and query.
  • GEO doesn't yet have a proven shortcut. Clear, accessible content and credible authority are still the most defensible foundation.
  • Measure revenue and qualified outcomes, not rankings or impressions alone. Even a top position can produce fewer visits than it used to.

Conclusion

AI Overviews change the economics of search by giving users more information before they reach a website. The best available experiment suggests this can materially reduce outbound clicks, but the size of the effect varies - and the broader traffic picture has limits. GEO may help your work become a cited source, but it isn't a proven substitute for visits lost from search.

Keep your SEO basics solid, publish content people can trust and actually use, and watch where your audience encounters your brand. Most importantly: connect search activity to business outcomes. When clicks are less predictable, the useful question isn't only whether you rank - it's whether the visibility you earn translates into revenue, retention, or whatever outcome matters most to your business.

Frequently Asked Questions

A randomized US desktop experiment found about 40% fewer clicks to external websites on searches that would show an AI Overview. Across all searches in the study, the estimated loss was about 18.5% of organic clicks. The result is strong causal evidence, but it does not predict the exact impact on every site or device.
GEO focuses on whether AI-generated answers cite or use your content. Current evidence suggests it relies mostly on SEO fundamentals, plus being present in sources AI systems draw on and making pages clear and easy to quote. There is no established set of technical GEO tricks that reliably produces citations.
There is no convincing evidence that llms.txt increases citations. Large observational studies have found no measurable relationship, and Google says it does not require AI-specific files for its search features. Maintaining one is unlikely to be harmful, but it should not take priority over accessible, useful content.
No. Search remains a much larger traffic source than AI assistants in the available multi-site data, and strong SEO helps users and systems discover your pages. The change is in how you evaluate results: track qualified visits, conversions and revenue, because rankings and impressions no longer guarantee the same click volume.

Related articles