Eric Van Buskirk

Eric Van Buskirk

@ericvanb.bsky.social

Founder of ClickStream Solutions. We turn search and AI data into headline-ready research. From study design and analysis to insights your audience will act on. AI Bard-Ass. https://clickstream.cc

577 followers275 following154 posts

Pinned

I analyzed, interpreted, and produced findings on cursor movement and scroll behavior across 846,000 Google Search sessions. Clickstream data provided by Surfer SEO. 📷 This is the layer that other clickstream data does not show.

In our Google AI Mode study for Citation Labs, participants clicked nothing in 64% of 149 tasks. In our ChatGPT study for Profound, 205 of 221 tasks (92.8%) ended without a click to an external site. They compared brands and picked one. Your analytics show none of it.

We analyzed 221 screen-recorded ChatGPT tasks from 56 U.S.-based regular users for Profound, combining behavioral coding, think-aloud transcripts, and questionnaires. We examined who defined the evaluation criteria: the user, ChatGPT, or both.

Looks fantastico. Rustling up some friends to read with beach campfire: reading aloud as part of scary-story night.

kate conger@kateconger.com

Hello! If you're new here, I write about Elon Musk, X, and recently, DOGE for the New York Times. I'm always interested in speaking to people who have worked with Musk, at X or in government, and I just verified my Signal on my NYT bio so you can contact me securely. www.nytimes.com/by/kate-conger

I analyzed, interpreted, and produced findings on cursor movement and scroll behavior across 846,000 Google Search sessions. Clickstream data provided by Surfer SEO. 📷 This is the layer that other clickstream data does not show.

In the Clickstream Solutions AI Overviews data study, I sampled cursor position every second, up to 60 times per session, capturing X-Y coordinates, viewport size, and a timestamp. Stitched together, that builds a spatial-temporal trail.

In my Google “scroll and hover” study, users with an AI Overview kept their cursor nearly still 44.1% of the time, versus 28.6% without one (+54%). At the same time, their cursor traversed 83.4% of the visible search page, compared to 66.2% without an AIO (+26%). I call that the cursor paradox.

View full thread
View full thread

9/ The practical output: this is no longer a ranking problem. It's a comparison problem. 🔁 Title tags and meta descriptions are now competing assets in a slower, more deliberate decision environment.

View full thread

6/ He named something the data implies but my analysis didn't make explicit: the second impression is when comparison happens. Your listing may be seen 2 or 3 times in a single session. The return visit is when users actually decide.

7/ Other ideas: Connecting backscrolling to his May 2025 qualitative study, where 38% of AIO sessions included reassurance-seeking clicks to a second source. Validation has moved onto the SERP itself.

View full thread

4/ AI Mode users accept the shortlist and move on. AIO users pause, spread across the page, scroll back up, and reconsider. His Netflix browse metaphor captures that contrast better than any stat. 🎬

5/ My original framing: AI Overviews are making Google less like a fast pass-through and more like a place where users read, revisit, compare, and reconsider. Kevin added a very useful layer to that.

It’s live. 🔍 846,000 Google Search sessions. Scroll depth. Cursor movement. Session retention. Backscrolling. All measured across search intent types, with and without AI Overviews. 🧵

These winners Trump hired are some kind of Team-Einstein, eh? “You can’t do a nuclear thing in 72 hours on the back of a napkin,” Mr. Rubio said in a brief interview Sunday.

I analyzed, interpreted, and produced findings on cursor movement and scroll behavior across 846,000 Google Search sessions. Clickstream data provided by Surfer SEO. 📷 This is the layer that other clickstream data does not show.

64% of AI Mode tasks, users made a product decision without clicking a single link or leaving the interface. Read. Evaluate. Decide. No tabs opened. AI Mode doesn't remove the comparison phase. It absorbs it into the answer. See the study I did with Citation Labs. Link 👇

Since click data can't tell us what people do inside a walled garden, how do you know what they DO? That's what I've focused on for a year now. Build user panels. Get their approval for desktop recording. It's not about what they SAY they do. It's real tasks, real activity.

1980s family. When I was too cool for school, my sister’s hair was too big for anything less than a station wagon, and my cousins were too preppy to drive south of Connecticut.

As someone who's lived in Vietnam for much of the last ten years, I know well what a surveillance state looks like. I know authoritarianism. I know the particular flavor of performance that happens when powerful people decide it's better to flatter a strongman than stand on principle.