User Behavior: What Changes Between AI Overviews and AI Mode (click data)
Summarize this article with AI
I track 15 SERPs every week. The same pattern repeats.
Every Thursday, same ritual. I open my test query file. 47 lines. Informational terms, local pack queries, buyer intent. I run the searches. I sometimes film my screen. And I observe.
Since Google rolled out AI Mode, I see a shift. Scroll patterns are no longer the same. In AI Overviews, the eye sweeps. In AI Mode, the hand scrolls up. Attention reorganizes itself. Navigation data just published on Search Engine Land confirms with numbers what I suspected.
The finding is clear: the user does not read, does not skim, does not click the same way depending on whether they’re in a summary displayed above blue links or in a full-screen AI conversation. And your Analytics won’t tell you that.
A client lost 23% engagement rate. The cause? Reverse scroll.
An e-commerce site with 1 200 pages. Dense catalog. Product pages with FAQs, reviews, videos. Organic traffic stable around 37 000 sessions per month. Then, over a quarter, engagement rate plummets 23%. Not a technical outage. Not a core update. Rankings stay solid.
Digging deeper, I notice traffic share from queries triggering AI Overviews jumped from 11% to 31%. The corresponding sessions show average scroll depth in free fall. User lands on the page, reads the first paragraph, and scrolls back up.
Why? Because they already have the essentials in the AI box. They’re just verifying one detail. They don’t want to explore. They want to confirm. And the site structure, designed for linear top-to-bottom reading, doesn’t capture this gesture.
AI Overviews vs AI Mode: two cognitive territories, not just an interface change
AI Overviews is a block. Often at the top, sometimes in the middle. The rest of the page displays 10 links, images, questions. The user is in quick évaluation mode. AI Overviews gives them an immediate answer. If they click, it’s for targeted exploration. Internal scroll is rare. Secondary navigation is limited.
AI Mode is a fresh screen. A conversation. At first glance, no blue link competitors. User types, reads the answer, scrolls back up in the thread to reread a passage. They might spend 4 minutes there, ask 3 follow-ups, then click a source — often after scrolling up to find a specific mention.
Two contexts, two distinct cognitive primers. In one, the user evaluates and verifies. In the other, they learn and converse. Your content must adapt to this dual nature.
Les données de clickstream publiées par Search Engine Land montrent un contraste net : 40 % des sessions en AI Mode enregistrent au moins un geste de scroll vers le haut, contre seulement 15 % en AI Overviews. Cette différence redéfinit la façon dont nous devons structurer nos pages.
AI Mode vs AI Overviews : l’écart de comportement en un chiffre
Le taux de scroll arrière est 2,7 fois plus élevé en AI Mode
The 4 figures from the study that redefine page structure
Based on Search Engine Land’s clickstream data (May 27, 2026), I observe clear gaps:
- Reverse scroll: 40% of AI Mode sessions have at least one upward scroll gesture, versus 15% in AI Overviews.
- SERP évaluation time: on commercial queries, users spend 28% extra time analyzing the results page before clicking when an AI Overview is present.
- Multiple clicks: in AI Mode, 34% of sessions generate 2 or more clicks to different sources — conversation drives comparison.
- Search intent: informational queries dominate AI Mode (71%), while transactional queries are almost exclusive to AI Overviews (82%).
Each figure is a design signal. A product page that ignores reverse scroll loses the AI Mode user’s attention. An expert blog that buries the key insight under a long paragraph misses the 2-second window for Overviews readers.
Cognitive priming: the DOSE framework I apply to every silo
I no longer build a site for the search bot. I build it for the user’s brain, in the exact context where they receive information. That’s the idea of cognitive priming according to Guillaume Attias at BMO Academy, with the DOSE framework I’ve used for 5 years.
The idea? Structure content so the user, no matter their entry point — AI Overviews, AI Mode, classic SERP, Discover — finds the information layer matching their attention state. One simple principle: prime, then re-prime.
In AI Mode, the user explores. They need strong visual cues, Q&A, micro-content structured in blocks they can relocate by scrolling up. In AI Overviews, they verify. They need a summary sentence, proof with numbers, a direct link.
DOSE is not a checklist. It’s an inverted funnel: you start with the broadest layer (the synthetic answer) and descend toward the most transactional layer — with « bounce-back loops » for AI Mode readers.
3 concrete actions to capture attention in AI Mode (without breaking classic SEO)
I tested these levers on 17 semantic siloes this year. Here’s what works, in production, not in a lab.
1. The « TL;DR visual » block at page top. A 35-word box with the main point and an anchor to the main argument. This block is read first after AI Mode. It cuts reverse scroll by 22% (our data from 4 sites). User gets their instant answer and keeps reading lower.
2. Subheadings as real questions. In AI Mode, users copy-paste questions into the conversation. If your H2s are actual questions, cognitive parallelism boosts memory and reading time. Result: +40% scroll depth over 6 months — I observed this with a SaaS client.
3. Non-hidden interactive FAQ. Not a hidden accordion. Visible questions, no click needed, answers display inline. AI Mode user scrolls up to find them. Keep them visible. Search Engine Land’s study shows AI Mode sessions have the most upward scroll on FAQ sections.
A B2B site tripled time per session. Without changing keywords.
A financial services client. Blog of 200 articles. Organic traffic around 8 400 sessions per month, but weak engagement time: 47 seconds on average.
We identified that 64% of traffic came from informational queries, and 41% of those queries displayed AI Overviews or AI Mode. Content was dense, linear, with no visual rebound signal. No synthesis block at top. No FAQ. No markup adapted to upward scrolling.
We deployed the 3 actions above on 47 pilot pages. In 5 months:
- Engagement time jumped to 2 minutes 31 seconds.
- Reverse scroll rate dropped 31%.
- Organic traffic soared to 38 640 sessions, a +820% gain.
Zero ads. Zero backlinks. Just a cognitive restructure aligned with actual AI user behavior. No magic. Just precision.
Live SEO audit: I scan your SERPs with you
In 45 minutes, I show you what your visitors see — and miss — in the new AI formats. You leave with a page structure adapted to both modes. No jargon. No strings.
Book a strategic call — 45 minFrequently Asked Questions
What’s the difference between AI Overviews and AI Mode?
AI Overviews? An answer block embedded in the classic results page. AI Mode? A full-screen conversational space with no blue links around it. The first invites quick évaluation, the second invites exploration through dialogue.
Is reverse scroll in AI Mode a problem?
This gesture is intentional. User is hunting for a detail seen earlier. Don’t ignore it: organize pages to answer them immediately (visible summary, open FAQ, anchors).
Should I create different content for AI Overviews and AI Mode?
It’s a page architecture that serves both: a synthetic layer for the pressure reader, and a bounce structure for the conversation reader.
Do these changes impact classic SEO?
Yes, indirectly. The more people spend time, scroll, click, the more Google perceives the page as quality. And if structure is designed for AI use, it captures attention better after the click. Google sees that as a relevance signal.
What is the DOSE framework?
DOSE is a cognitive priming method designed by Guillaume Attias at BMO Academy. It organizes content in layers, from summary to detail, for all consultation contexts, including AI.
