64 people got wired up to watch them scroll.
What they felt predicted nothing.
A camera read their faces. Electrodes read their skin. Then researchers handed them a phone: a simulated Instagram feed, twenty-nine posts, eight sponsored and twenty-one organic. Close to the ratio your own feed runs at.
The baseline first. Scrolling itself barely registered as anything. In the study's own words, browsing was "predominantly neutral in affective tone, interrupted by brief spikes in arousal and positive valence."
Sponsored posts did register. Faces showed less neutral expression on sponsored content than on organic content: 68.08 versus 89.12, a statistically significant gap. Something changes when an ad hits the screen. People do feel it.
That measurable reaction didn't predict whether anyone engaged. The paper's own line: "biometric indicators of emotional arousal and valence did not predict engagement behavior."
A reaction happened. Whatever decided to tap or keep scrolling never consulted it.
The study doesn't tell you what does predict engagement. It tells you what doesn't. Researchers tested that mechanism directly: sixty-four people, facial coding, skin-conductance readings across dozens of trials each. Not a survey. Not a focus group's guess about how they felt.
If you're still testing which version of an ad made people feel more, you've been measuring the wrong number.
Sources:
Frontiers in Neuroscience (2025) — "Disrupting the browsing experience: impact of sponsored social media content on affective flow without driving engagement"
https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1636848/full
Cogent Business & Management (2026) — "Beyond emotion: exploring image memorability with eye tracking," covered by Phys.org
https://www.tandfonline.com/doi/full/10.1080/23311908.2026.2677960
https://phys.org/news/2026-07-visuals-online-ad-emotion.html
Featured image: Xtrodes