The average Netflix user spends 18 minutes browsing the home screen before picking what to watch. They scroll past tiles, hover for trailers, scroll back to a show they almost picked, then circle back to the row they started in. The browse is the experience.
Search isn’t that different anymore. We just lacked the data to see it until now.
This week:
- Four behavioral shifts that show up when an AI Overview is on the page, measured across 846,000 real Google sessions.
- Why brand-name searches no longer give you the shortcut they used to.
- One finding that should change how you write title tags and meta descriptions this quarter.
Eric Van Buskirk of Clickstream Solutions analyzed anonymized clickstream data provided by Surfer SEO, drawing findings from approximately 846,000 U.S.-based Google search sessions collected in February and March 2026.
It’s the fifth user-behavior study on Google’s AI features in the last 12 months. The 70-user UX study from May 2025 used think-aloud and screen recording. The 250-session AI Mode study from October 2025 captured how users behave inside AI Mode itself. This one trades qualitative depth for the scale to find behavioral patterns the smaller studies couldn’t see.
For context: Prior public Google SERP mouse-tracking studies measured dozens of people. The largest had a few thousand tasks. This study analyzed queries from a panel of tens of thousands of Google Search users.
The most significant pattern shown: Users behave in opposite ways in AI Overviews and AI Mode. AI Mode is autoplay. AI Overviews is the Netflix browse.
This article covers four findings from the new study (full methodology here) and what they mean for how you write title tags and meta descriptions in 2026.
Complete brand visibility: Whether you’re a global brand or nationwide nonprofit
Good Guys SEO secured buy-in and built an end-to-end search strategy for one of the US’s most recognized nonprofits using Semrush for Enterprise.
The result: +37% AI share of voice for homelessness services, with visibility connected directly to campaign traffic and donations.
Complete SEO and AI search capabilities in one platform. The outcomes speak for themselves, whatever your industry or business model.
1. Same user, opposite behaviors
In AI Mode, users often accept the answer while continuing to evaluate the SERP in AI Overviews.
The April 2026 study of 185 high-stakes purchases found that in 88% of AI Mode tasks, users took the AI’s shortlist as-is, 74% picked the item ranked first, and 64% clicked nothing at all. AI Mode often behaves like a closed loop: The user reads the answer, picks from inside the answer, and moves on.
Note: N/A means the behavior doesn’t apply to that surface. AI Overviews don’t deliver shortlists like AI Mode does, and cursor/scroll metrics were only measured in AI Overview comparison environments.
New cursor tracking data on AI Overviews shows a different behavior pattern. Cursor positions spread across the equivalent of 83% of a viewport, compared with 66% when no AI Overview appeared. Users kept their cursors still 44% of the time, compared with 29% without an AIO. And in the median session, reverse-direction scrolling accounted for nearly half of scroll movement.
Taken together, this suggests that AI Overviews turn the SERP into more of a comparison environment. The user reads, pauses, weighs, returns to earlier content, and reconsiders before clicking.
The implication for strategy: AI Mode optimization and AI Overview optimization are not the same job. Showing up in an AI Mode shortlist is a visibility problem at the model layer. Showing up in an AI Overview is a comparison problem on the SERP itself.
Among users who reverse direction in an AI Overview SERP, the median user spends 47.5% of total scrolling going back up the page. Without an AI Overview present, that figure is 27%.
The cursor data shows scrolling isn’t a one-way trip. On AI Overview SERPs, scrolling is roughly a 50/50 split between going down and going back up. That’s the behavior of someone re-reading, not someone scanning.
The 70-user study from May 2025 found this pattern qualitatively. 38% of AI Overview sessions in that study showed reassurance-seeking clicks where users opened a second link “just to be sure.” The new clickstream data shows that the validation work has now moved onto the SERP itself. Users used to leave Google to validate a result. Now they validate by reversing back over results they already saw.
This is exactly the Netflix browse pattern. You hover on a tile. You scroll past it. Something pulls you back, so you scroll up to read the description again. The decision happens during the reversal, not the first pass.
For e-commerce and high-consideration decision categories, this is the most consequential finding in the study. Your listing in an AI Overview SERP isn’t getting one impression anymore. It’s getting 2 or 3, and the second impression is when comparison happens.
3. Search type no longer predicts behavior
For 2 decades, search intent has been the foundational segmentation framework in SEO. You could predict how long a user would stay on a Google SERP by knowing what kind of search they did. The cursor data shows that’s no longer true when an AI Overview is on the page.
At 21 seconds into a session without an AIO, only 12% of navigational searchers are still on the page. 32% of local searchers are. In classic search, time-on-page has always followed intent: navigational users leave fast because they know where they’re going, local users stay because the SERP is dense with maps and listings, informational users fall somewhere in between. That 20-point spread is what every SEO mental model is built around.
With an AI Overview present, the spread compresses to barely 6 points. All five intent types (informational, local, navigational, transactional, video) cluster between 41.9% and 48.5% time-on-page at 21 seconds. Intent stops predicting how long users will stay in the search results.
This is the most novel finding in the study. Nothing in prior research has shown AI Overview collapses time-on-page differences between intent types into a single band. The 70-user UX study segmented by query risk, the 250-session AI Mode study by task type. Both showed intent matters for engagement. The new clickstream data says intent stops predicting how long users stay on the SERP once an AI Overview appears.
The scoping matters. This is a time-on-page finding, not a “behavior” finding. Scroll depth still varies by intent under AIO, and it actually reshuffles: local jumps from third deepest to first, video falls from first to third. That part of the story is in the next section.
The practical implication is uncomfortable. Most SEO playbooks recommend different optimization patterns by intent. Local pages get one treatment, transactional pages another, informational pages a third. However, our time-on-page data suggests that when an AI Overview is on the SERP, users stay for similar amounts of time regardless of why they searched.
The main takeaway here is that intent-based segmentation still matters for what content you write, but it matters less for predicting how long users will stick around on the SERP itself.
4. Brand searches lost their shortcut
Cursor activity for navigational queries (someone typing a brand name into Google) increased 40% when an AI Overview is on the page. Even users who came knowing where they wanted to go are sweeping the page first.
Without an AIO, navigational searchers were the most focused group in the study. They scored 19.7 on the cursor scatter measure (essentially, how much people bounce around the page), the lowest of any intent type. Only 12% were still active at 21 seconds. They came to Google with a destination, found it, and left.
With an AI Overview present, that profile changes completely. Cursor scatter for navigational searchers jumps to 27.5. 45.8% are still active at 21 seconds. The brand-name shortcut, the fastest path through Google search for the past 20 years, no longer works the way it used to.
The 70-user study from May 2025 found that brand and authority is the first gate users apply when reading a SERP. They check who’s cited before they check what’s said. The new clickstream data shows that the gate now fires even when there’s nothing to gate against. A user who typed “Lenovo” is still doing the authority check on the AI Overview content first.
This is the same Netflix pattern again. You open a show you’ve already decided to watch, but you pause first to read the synopsis and check the rating. The brand recall got you to the page. The browse decides whether you click through.
The implication for branded search: Brand recall is no longer enough on its own. Even users who searched for you specifically are now evaluating what’s around you on the SERP before they click through.
This post first appeared on the author’s website and is republished here with permission.
Contributing authors are invited to create content for Search Engine Land and are chosen for their expertise and contribution to the search community. Our contributors work under the oversight of the editorial staff and contributions are checked for quality and relevance to our readers. Search Engine Land is owned by Semrush. Contributor was not asked to make any direct or indirect mentions of Semrush. The opinions they express are their own.