What Would GPT Click: Practical Effects of Human-AI Behavioral Misalignment and the Cost of Synthetic Participants in User Experience

TL;DR

This study systematically evaluates GPT's ability to mimic human click behavior and reasoning in UX tests, revealing significant discrepancies and limitations that impact decision-making accuracy.

Contribution

It provides the first comprehensive analysis of the practical issues and distortions in synthetic user responses generated by GPT in real UX research contexts.

Findings

GPT significantly differs from real user data in 53% of tasks

Efforts to improve fidelity via personas and reasoning methods are ineffective

Synthetic responses exhibit distortions that undermine their usefulness for UX decision-making

Abstract

Synthetic participants represent a methodologically concerning concept that threatens the integrity of UX research. Findings from previous experiments specify how AI outputs are misaligned with the behaviors and thoughts of real humans in various ways. However, industry voices keep underestimating their severity, advocating for practical compromises where good-enough data does not need to be perfect, and all issues will be solved by future tuning. Our study tackles the lack of systematic understanding of the practical issues that arise with synthetic behavior and its use for steering decisions within real contexts. Within twelve diverse first click tests (n = 3431) obtained from real UX practice, we examine the ability of GPT to predict where humans click and how they reason about their behavior. Results (e.g., significantly different distribution from real data in 53% of tasks) demonstrate critical failures to reflect the patterns in which users click on visual elements and the underlying cognitive processes. Participant personas, chain-of-thought reasoning in GPT, and different sampling parameters fail to create sensible fidelity improvements apart from inflating believability. We expose a multitude of nuanced distortions in synthetic responses that reduce their overall analytical usefulness as a decision-making resource, compared with real data. Observed distortions can be theoretically linked to the properties categorically inherent to LLMs: their statistical nature and encoding of semantic heuristics dependent on their training on linguistic data.