Can ChatGPT Reflect Professional HACCP Judgments? A Comparative Study in Hospitality Food Safety

The implementation of Hazard Analysis and Critical Control Points (HACCP) systems remains a cornerstone of food safety management. However, their effectiveness is influenced by organizational, human, and technological factors. This study investigates professional perceptions of HACCP implementation and explores the extent to which a large language model (LLM)-ChatGPT 4.1-can approximate human judgments in this domain. A structured questionnaire was administered to 90 professionals operating in food service, hospitality, industry, and consultancy. Responses were compared with outputs generated by ChatGPT 4.1 via a standardized multi-persona prompting protocol simulating five professional roles. To enhance response stability and minimize stochastic variation, each question was submitted in independent zero-shot sessions over multiple iterations. Responses were analysed across three thematic dimensions: barriers to HACCP implementation, perceived benefits, and digital readiness. Spearman correlation analysis of human responses revealed a systemic "training-turnover association," where high staff turnover (r = 0.62, p < 0.01) was significantly associated with difficulties in maintaining continuous training. Statistical benchmarking using one-sample t-tests showed that the tested model generated significantly higher ratings for implementation barriers (p < 0.001) and digital readiness (p < 0.001) than the corresponding human assessments. These findings suggest that while ChatGPT can approximate aggregated professional perceptions in certain areas, notable divergences persist in operational and readiness-related domains. The study contributes methodological insights into human-AI comparative research and highlights opportunities and limitations of AI-supported decision-making in food safety management systems.