Four online experiments found that users' beliefs about a source's trustworthiness biased their decisions by up to 30% — more than the information the source actually provided.
We primed one group to see a source as trustworthy, and another to see the same source as suspicious. Both groups then received the same information from it. The trusting group followed the information; the suspicious group didn’t — a gap of about 30 percentage points.
Anyone building search, recommendations, reviews, or credibility signals faces a version of the same question: when users judge a source, do they update on what the source actually shows them — or are they mostly acting on a prior belief about whether the source can be trusted? The answer determines how much weight a product should put on credibility cues versus content. It matters for SERP ranking, review systems, expert recommendations, and content-credibility labels alike.
I ran a four-study program. The first two were online A/B experiments (N=157, within- and between-user) built in Python, where users interpreted on-screen search behavior after being primed to see a source as trustworthy or not. I analyzed the data with mixed-effects models in R to account for repeated measures and participant variance that a simple ANOVA would confound.
I then extended the design in two further experiments (N=360) — reported in a companion paper now under peer review — that segmented users by their level of sub-clinical neurodivergent traits, testing whether the trust effect held uniformly or whether different user groups updated their trust differently. The replication-plus-segmentation logic is what turns a single finding into an actionable, generalizable one.
Gerlofs, D. J., Roberts, K. H., & Kingstone, A. (2025). Perceived intent drives gaze interpretation. Attention, Perception, & Psychophysics, 87, 2323–2331. https://doi.org/10.3758/s13414-025-03149-9
Trust framing shifted decision rates by roughly 30 percentage points. Crucially, the actual information the source provided moved decisions less than the framing around it did. And the effect wasn't uniform: neurodivergent user segments showed distinct trust-updating patterns — meaning a single trust-signal design can serve some users well and others poorly.
The findings have direct implications for how products frame credibility: source badges, ranking positions, and trust cues can move users more than the content they label. The segmentation result points to concrete inclusive-design changes for neurodivergent users. The core study is published in Attention, Perception, & Psychophysics (2025); the neurodivergent extension is under peer review.
If a credibility label or ranking position shifts decisions more than the content it labels, that matters for how products communicate information, handle trust signals at scale, and avoid biasing users in ways that are hard to audit — and it’s rarely measured. Because different user segments update trust differently, a single trust-signal design is rarely neutral across a whole user base.