Using LLMs in sensory service research: initial insights and perspectives

Abstract

Researchers have started using large language models (LLMs), such as OpenAI's GPT, to generate synthetic datasets designed to mimic human response behavior. Several studies have systematically compared LLM-generated data with human samples in order to explore LLMs’ ability to mimic consumer decision-making. Extending prior findings, our research sets out to explore how GPT-4o responds to sensory information, and to evaluate its ability to grasp crossmodal correspondences as well as multisensory congruence–as commonly encountered in service settings. Our results indicate that while GPT-4o identifies and describes sensory stimuli accurately, it often fails to replicate the associative meanings and interpretations that humans derive from these stimuli, especially in stand-alone assessments. Our research therefore underscores the need for further exploration of the conditions under which LLMs reliably mirror human responses to sensory stimuli, and the implications of using LLMs in research on sensory-rich service settings.