A comparison of human and LLM-simulated participants in a writing style task

2026-06-15

AI Summary

This study investigates whether large language models (LLMs) can effectively simulate human participants in research on writing style preferences, particularly regarding their suitability for evaluating human–computer interaction systems. By systematically comparing the performance of 30 real users with that of GPT-4o as simulated participants in a dynamic preference inference task, the work provides the first comprehensive analysis of both agents’ ability to adapt to individual preferences and maintain behavioral consistency across repeated interactions. The findings reveal that GPT-4o exhibits insufficient depth and inherent biases in textual judgments, hindering its capacity to accurately replicate the evolution of human preferences. Moreover, the study uncovers substantial preference inconsistencies among human participants themselves. These results prompt methodological reflection on the use of LLMs in simulation-based experiments and underscore the continued necessity of human involvement in automated evaluation frameworks.

Abstract

Because large language models (LLMs) can produce natural language that is sometimes indistinguishable from texts produced by people, some researchers are starting to consider replacing human participants with LLM simulations. In this study, we test the extent to which the findings of a simulation with an LLM prompted to act as a synthetic participant match those obtained from 30 human participants. In our experiments, we evaluated how well writing style preference inference algorithms adapted to a participant over repeated interactions, compared to a baseline. We discover hints of bias and a lack of depth in GPT-4o's text generation and judgement that prevent it from accurately simulating people's behavior. Our results also hint at human biases that highlight the importance of considering human factors in the evaluation of systems that depend on human-automation interaction. Rather than treating these discrepancies as evidence for or against the validity of LLM-simulated participants, we present this study as a case analysis of methodological and design challenges.

Problem

• large language models
• human simulation
• writing style
• participant behavior
• human-automation interaction

Innovation

• LLM-simulated participants
• writing style preference
• human-AI interaction
• behavioral simulation
• methodological challenges