Bridging the gap between human-based and AI-based food perception: A comparative study of sensory preferences across gender and age groups

With the growing integration of artificial intelligence into consumer research, this study explores the use of ChatGPT to simulate food-related preferences. The primary goal was to assess how closely AI-generated responses reflect real human sensory choices. A total of 240 real participants were recruited across three age groups (under 10, 20–30, and over 40 years), equally distributed by gender. In parallel, 240 artificial consumer profiles were generated in ChatGPT. Participants completed a visual questionnaire covering four domains: preferred potato shape, preferred frying method, the impact of nutritional information, and willingness to pay more for healthier products. Results showed that both human and AI groups preferred thick-cut fries and deep-frying. However, when nutritional information was provided, a noticeable shift toward air-frying occurred, particularly among women and individuals over 40. Overall, the AI responses aligned with human data in over 80 % of cases. That said, discrepancies became apparent in questions requiring higher-level reasoning, such as interpreting nutritional labels or making economic trade-offs. The reduced accuracy in complex decision-making contexts indicates that these models require further refinement and validation before they can be considered completely reliable alternatives to human-based studies. Despite these limitations, the findings show that language models hold strong potential for use in early-stage sensory evaluations and consumer-related decision-making, a valuable capability particularly when access to real participants is limited, such as during pandemics. These insights could support the development of innovative food products that better align with the preferences of diverse consumer groups.