TL;DR: Google AI Overviews reduce click-through rates by 30-60% on affected queries. Here's the data and how to adapt.
Picture a search results page where just one extra panel — the pastel-boxed AI Overview — causes nearly half of your potential visitors to vanish. That's the reality Pew Research Center surfaced when it tracked 900 real users in March 2025. On a conventional Google page, about 15 percent of people clicked through to a website. Add an AI Overview, and the click-through rate cratered to 8 percent.
The math is brutal: for every thousand searches, you've just lost seventy clicks — eyeballs that would have landed on your articles, storefronts, or sign-up pages. Worse, Pew found that the AI tile itself rarely drives traffic back to sources. Barely 1 percent of summaries produced a click on a citation, and when they did, the winners were the giants — Wikipedia, YouTube, Reddit — leaving niche publishers, SaaS docs, and indie blogs out in the cold.
Here's what this means for our business specifically: SEOJuice helps sites get more organic traffic. When the primary mechanism for discovering content — clicking a search result — shrinks by half on 20% of queries, that's not an abstract problem for us. It's existential. We've spent the last six months rethinking what "optimization" means when the search result page itself answers the question. Some of what we've learned is in this article. Some of it we're still working through.
Google's AI Overviews sit atop the results page like a super-charged featured snippet, pulling text from multiple sites and generating a conversational answer. The feature began as Search Generative Experience (SGE) in May 2023, visible only to volunteers in Google's Labs. Over the next twelve months, the company expanded SGE, testing follow-up question chips, inline shopping links, and citation cards.
In May 2024, Google dropped the "Labs" label and promoted AI Overviews into the default U.S. search interface — initially on a narrow set of product and how-to queries. Each month afterward, the coverage widened: recipes in August 2024, health questions in November, travel planning in early 2025. By July 2025, BrightEdge and other tracking firms estimated AI Overviews appeared on roughly 20 percent of all desktop searches, with the likelihood soaring to 60 percent for question-style queries. The AI tile has become a second front page — one that often answers before users ever reach the blue links.
To quantify the impact, the Pew Research Center partnered with Ipsos's KnowledgePanel, equipping 900 U.S. adults with a browser plug-in that logged every Google search they ran in March 2025. The plug-in recorded:
Query text and length (single word, phrase, or full question).
Presence or absence of an AI Overview on the resulting SERP.
User interactions — scroll depth, link clicks, and whether the session ended on the SERP.
Pew then compared click-through behaviour on pages with versus without the Gemini-generated panel. Their headline finding: pages that showed an AI Overview cut outbound clicks nearly in half — from 15 percent down to 8 percent. Even more striking, only 1 percent of all AI Overviews produced a click on any cited source.
Google's PR team dismissed the study, arguing the sample size and "skewed queryset" misrepresent overall traffic. Pew counters that its opt-in panel methodology is identical to the one researchers have trusted for election polling and media-consumption studies for two decades. I find the Pew data directionally convincing, though I'd love to see a larger-scale replication. 900 participants is meaningful but not definitive. What makes me trust it more is that the numbers are consistent with what we see across our own customer base — sites in informational niches have reported 25-45% drops in organic CTR on queries where AI Overviews appear.
Pew's dataset turns a year of speculation into hard numbers. The headline metric: when an AI Overview appears, the likelihood of any organic click on that SERP falls from 15 percent to 8 percent — almost a 50 percent haircut.
The promised "traffic dividend" from citations barely exists. Out of thousands of AI tiles logged, just 1 percent produced even a single click on a cited source link. Ninety-nine times out of a hundred the AI answer satisfies (or stalls) the user before a publisher can claim so much as a page-view.
Question-style searches — the bread-and-butter of informational content — are exactly where the robot intervenes: 60 percent of queries phrased as questions now trigger an AI summary.
And the spoils are concentrated. Three destinations — Wikipedia, YouTube, and Reddit — account for 15 percent of all citations the model makes, leaving the remaining 85 percent to be divided among every other site on the internet. That concentration bothers me. It suggests that being "good enough" won't cut it — you either make it into the model's citation inner circle or you're fighting over scraps.
Pew's citation log reads like a popularity contest skewed toward scale and community muscle. Authority encyclopedias (Wikipedia) and user-generated behemoths (YouTube videos, Reddit threads) dominate because they offer expansive, continuously updated content pools the model can trust statistically. Mega-publishers with DR 90+ — think Mayo Clinic, The New York Times, and Investopedia — show up often enough to keep their brand equity intact.
The casualties sit at the long tail: niche newsrooms, specialist SaaS blogs, academic journals behind paywalls, and local outlets. These sites rarely appear in the AI panel, and when they do, users almost never click through. Even when a smaller publisher provides the original scoop, Gemini often paraphrases it and attributes to a higher-authority roundup or a crowd-sourced forum.
For marketers, the takeaway is clear. If you're not already in the citational "inner circle," relying on traditional blue-link SEO alone won't protect your traffic. The strategy frontier shifts to structured data, concise answer blocks, and explicit invitations for AI crawlers — all aimed at convincing the model you deserve one of the few outbound links it still doles out.
Google's new interface rewards sites that feed its models clean, verifiable data — and quietly sidelines the rest. Three priorities now separate the winners from the traffic orphans.
1. Welcome the Bots That Write the Answers
Blocking GPTBot, ClaudeBot, Perplexity-Bot, or Google-Extended is the modern equivalent of adding noindex to your homepage. Leave reputable AI crawlers unmolested in robots.txt (User-agent: * Allow: /) and disable any "Block AI Scrapers" toggle in Cloudflare or your firewall. The crawl cost is pennies, but every successful fetch becomes training fuel that can surface your paragraph — complete with a link — in the next billion chat queries. We wrote a detailed guide on fixing the Cloudflare toggle if you're not sure how.
2. Serve a Fact-Checked Answer Box on Every Page
AI Overviews pluck concise statements, numbered steps, and bullet-point summaries. Give them exactly that: a 40- to 60-word answer block near the top of informational articles, wrapped in an <h2> like "Quick Answer." Make it verifiable — cite the study, include the year, link to the primary source — and keep promotional fluff out. This is where I'm genuinely uncertain about best practices. We've tested this on about 40 client pages and seen mixed results — 60% showed improved snippet capture, but the other 40% saw no change. The data is promising but not conclusive.
3. Double-Down on Brand Queries and E-E-A-T
If generic clicks are drying up, protect — and expand — the traffic you own outright: searches that include your brand name. Optimise title tags and meta descriptions to pair the brand with core topics. Bolster E-E-A-T by adding author bios with credentials, linking to peer-reviewed citations, and maintaining a transparent "Last Reviewed" timestamp. The stronger your perceived authority, the more likely Google's AI will quote you instead of a forum thread.
Pew's numbers are the canary in the SERP: clicks are evaporating wherever AI answers appear, and the coverage area is only widening. Sites that cooperate with reputable crawlers, package their expertise in citation-ready formats, and fortify brand authority will still earn visibility — both in traditional blue links and inside the AI tiles themselves.
Everyone else risks becoming background training data, remembered by the model but never visited by the user. That's a strange new form of being famous and invisible at the same time. Adapt your SEO playbook to this AI-first reality now.
Related reading:
no credit card required
