In one sentence
On May 26, 2026, ClickStream Solutions (with Surfer SEO data) published an analysis of 846,000 U.S. Google sessions. The verdict: with AI Overviews present, 47.5% of scroll movement goes backward, the cursor spans 83% of the viewport (vs. 66% without), and 38% of sessions show a "reassurance click." The SERP is no longer a list — it's a comparison page.
On May 26, 2026, Eric Van Buskirk (ClickStream Solutions) published on Search Engine Journal an analysis of 846,000 U.S. Google sessions collected in February–March 2026 from anonymized cursor-tracking data provided by Surfer SEO and re-read by Kevin Indig at Growth Memo on May 25. Methodology: cursor sampled at 1-second intervals, up to 60 samples per session, across three subsets (74,848 balanced, 99,994 representative, 99,994 filtered).
Why this study matters: it's the first time anyone has measured actual user behavior on AI Overviews SERPs at scale with cursor-level granularity. Every previous analysis — CTR averages from Ahrefs, Sistrix, Similarweb — measured outputs (clicks), not the reading motion itself.
The numbers that break your current SEO models
1. Re-scroll: the signal of a comparison, not a scan
Without an AI Overview, 27% of scroll movement goes upward — the classic scan-and-leave behavior. With an AI Overview present, that number jumps to 47.5%. On navigational queries (where users type a brand name), it rises from 23% to 44%.
Translation: users no longer scroll down to find. They scroll down, re-read the AI Overview, scroll down to read the cited sources, scroll back up to verify a blue snippet, scroll down again. The SERP has become a comparison page.
"The search result preview carries more weight. Users pause over search snippets rather than reflexively clicking the first result. Your listing may get multiple looks, not one."
— Eric Van Buskirk, ClickStream Solutions, Search Engine Journal (May 26, 2026)2. The cursor spreads: the decision happens on the page, not on the click
Without AI Overviews, the user's cursor covers 66% of the viewport during the session (measured at 60 time points). With AI Overviews, it covers 83%. In parallel, the cursor stays still 44% of the time (vs. 29%) — users are reading carefully before acting.
This is the exact inverse of behavior we saw two years ago in the featured-snippet era. Users hovered and clicked. Now, they settle in.
3. The reassurance click: 38% of sessions open a 2nd link "to verify"
This is the most operationally useful finding for anyone who still wants click-through traffic. 38% of AI Overview sessions show a two-step trajectory: the user receives the AI's answer, then opens a link "to verify." Not to discover the information — to confirm it.
Direct consequence: the 42% drop in organic SEO traffic observed over the year doesn't translate into a complete disappearance of clicks. A meaningful share of clicks survives — but they change in nature. You no longer capture the "first-curious" visitor; you capture the "verifier."
What this means for your strategy, concretely
Before: optimize the title + first line of the snippet for a reflex click. Now: optimize the entire snippet, the visible URL, the sitelink, and the coherence with the AI Overview itself to capture a validation click. Three different levers, to be measured separately.
Priority 1 — Become the source cited INSIDE the AI Overview
If the user reads the AI's answer first, they discover you via the links cited within the overview, not via the organic results below. Google's Preferred Sources, just extended to AI Overviews, sharpens this point: sources a user has marked appear first in the citation, with a distinct label. Organic ranking loses meaning on those queries.
Action: structure every article with a direct answer in the first sentence (40–60 words), citable as-is by the model. Numerical data, named sources, simple syntax. That's exactly the method our AEO framework documents.
Priority 2 — Treat the blue snippet like a mini-landing page
With 83% of the viewport scanned, your meta description and visible URL are now read, not skimmed. Rewrite meta descriptions so they answer a concrete question and contain a number or recognizable entity. The sitelink (URL + sitename + breadcrumb) becomes a major trust signal — not a technical detail.
Priority 3 — Build the "validation page"
The reassurance click opens a second link after the user has already read an answer. The landing page doesn't need to tell the story from the start. It needs to confirm quickly, ideally with a data point the AI didn't surface. Top-of-page bullet list, primary source cited, honest counter-argument. No marketing fluff.
Priority 4 — Measure "validation micro-traffic"
In GA4 and Search Console, distinguish "primary traffic" (user arriving directly) from "secondary traffic" (user arriving after reading an AI Overview). Signals: short session (15–30 s), little scrolling, but a low rate of return to Google. That's a successful validation click. As of today, no tool segments this traffic automatically — a monthly manual audit is the best workaround.
What the study doesn't say
- The scope is U.S.-only. Data comes from American sessions (Surfer SEO). Behavior on Google.fr or Google.de may differ — AI Overviews is rolling out unevenly across European verticals. Re-test on a European sample before generalizing.
- The study doesn't isolate by screen size. The 47.5% re-scroll rate is likely amplified on mobile (smaller viewport = more mechanical scrolling). Desktop/mobile breakdown isn't published. If your audience is >70% desktop, expect a muted effect.
- No conversion measurement after the click. The reassurance click survives, but we don't know if it converts as well as the old reflex click. Internal e-commerce CRMs will reveal this over the next quarters.
- "AI Mode" effect is different from "AI Overview" effect. In full AI Mode, 88% of users take the AI's shortlist as-is, with no further click. The behavior described here only applies to SERPs with AI Overview displayed — not SERPs fully replaced by AI Mode.
The Cicéro take
This study confirms what we've been observing with our clients for six months: SEO is becoming a presence problem, not a positioning problem. Being in the top 3 isn't enough if your snippet doesn't invite validation, and your validation page adds nothing beyond what the AI already said. The work has shifted from optimizing the click to optimizing trust.
That's exactly the methodology Cicero Studio applies: GEO audit + editorial production + automated semantic linking, with a minimum quality score of 90/100 on every article. Agency-quality work, software-grade productivity — from €250 to €1,800 per month. If you want to know where your site stands on these new signals, our analysis of how the AI Overview CTR rebound is reshaping SEO playbooks is a good place to start.
Primary sources
- → Search Engine Journal — "846,000 Google Searches Reveal How AI Overviews Are Changing User Behavior", Eric Van Buskirk (ClickStream Solutions), May 26, 2026
- → Growth Memo — Strategic re-read of the study (Kevin Indig, May 25, 2026)
- → Surfer SEO — Provider of the anonymized clickstream data analyzed in the study
Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility — on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.
LinkedIn