Implementing generative artificial intelligence in a tertiary hospital health screening center

Abstract Health screening centers perform high-volume, repetitive, information-intensive tasks that may be well suited to generative artificial intelligence (GenAI), but safe implementation requires accuracy, privacy protection, and workflow integration. In this study at Seoul National University Hospital Gangnam Center, South Korea, from June 2024 to May 2025, we conducted workflow mapping, user-requirement analysis, prototype development, and usability evaluation. Workflow analysis identified repetitive, protocol-dependent communication tasks across the screening service line. Screening center personnel reported that demand for GenAI was concentrated on improving workflow efficiency and reducing workload rather than supporting clinical decision-making. They prioritized pre-test guidance and automated test result explanation, while highlighting concerns about output accuracy, user acceptance, generational barriers, and privacy. An institution-specific retrieval-augmented generation chatbot showed favorable preliminary usability and received higher descriptive rankings than general-purpose artificial intelligence models in completeness, safety, and overall preference. These findings should be interpreted as preliminary usability evidence and require validation in larger evaluations. Overall, the findings support the potential role of GenAI as a human-supervised, institution-tailored workflow support tool in health screening services.