Accepted · in pressJEP: General

Users often don't know when they're right — and that shapes what they do next.

An eye-tracking study paired with large-scale behavioural data found that users systematically miscalibrate their confidence — and that better-designed feedback loops reduce friction and distrust.

Role: Lead researcher Org: Brain, Attention & Reality Lab, UBC Methods: Eye-tracking + clickstream Status: Accepted, JEP: General
Confidence rarely tracks accuracy on its own — but feedback design can close the gap.

Most feedback design assumes users know when they’re right. They often don’t — and the gap between how confident they feel and how accurate they are is a source of friction that products rarely measure.

The problem

User confidence — the feeling of knowing whether a decision was right — underlies trust, friction, and a sense of agency. A user who doesn't know when they're correct will either over-rely on a product's outputs or under-trust them, and both are bad outcomes. This matters anywhere users must judge their own choices: decision-support tools, recommendation systems, dashboards, and consequential workflows.

The approach

I paired an eye-tracking experiment (N=40) — collecting behavioural data alongside trial-by-trial confidence ratings — with a large-scale clickstream/telemetry study (N=326). The eye-tracking data pins down the mechanism; the telemetry tests it at scale. I used LLM-assisted scripting for parts of the analysis pipeline — a practical example of an AI-augmented research workflow that sped up a large, repetitive coding task without compromising rigour.

Methodology

Design
Eye-tracking (N=40) + clickstream A/B (N=326)
Measure
Trial-level confidence ratings vs. accuracy
Analysis
Mixed-effects models, confidence–accuracy correlation
Tools
Eye-tracker, Python, R, LLM-assisted coding
Status
Accepted, JEP: General

Full APA reference

Gerlofs, J. D., & Kingstone, A. (in press). Metacognition and the dual function of social gaze. Journal of Experimental Psychology: General.

In press — a DOI link will be added on publication.

What we found

Users were often poor judges of their own performance — confidence did not reliably track accuracy. But designs that better aligned decision outcomes with confidence cues (calibrated feedback) reduced this gap, and with it the unnecessary friction and trust deficits that miscalibration produces.

Impact

Accepted in Journal of Experimental Psychology: General. The product implications land on feedback design, confirmation UX, and any system where users must judge their own decisions — the question being: where does confidence miscalibration cause observable product problems?

So what for product

Confidence miscalibration is a hidden source of friction. When users feel uncertain about decisions they got right, they second-guess, abandon, or seek excessive confirmation. When they feel confident about decisions they got wrong, they don’t course-correct. Both degrade experience — and, in consequential products, outcomes. Feedback that helps users accurately assess their own performance is an effective intervention that is rarely measured, because most analytics track what users do, not whether they felt right about it.