Abstract
<title>Abstract</title> <p>Background Effective anamnesis is a core clinical competency, yet training opportunities with real or standardized patients are resource-intensive, difficult to scale, and particularly limited in online learning environments. Large language model (LLM)-powered virtual patients (VPs) offer a promising alternative, but most existing evidence focuses on medical students, and the integration of automated feedback and its perception across other health professions remains underexplored. This study examined healthcare students' perceptions of an enhanced GPT-4-powered VP with integrated automated feedback on professional and communication competence. Methods A cross-sectional study was conducted at IU International University of Applied Sciences (January–March 2025) among undergraduate students from eight health-related Bachelor's programs, including nursing, physiotherapy, occupational therapy, speech therapy, and nutritional science. Participants conducted an anamnesis interview with a chatbot-based VP simulating a patient with brain hemorrhage and multimorbidity, then received automated feedback before completing an online survey. Usability was assessed with the Chatbot Usability Questionnaire (CUQ); response quality, realism, behavioral engagement, feedback quality, and training benefit were assessed using purpose-developed, pilot-tested scales (5-point Likert), analyzed via medians, Cronbach's alpha, and Positive Response Proportions. Open-ended responses underwent qualitative content analysis following Mayring. Results Sixty students participated (83.3% female; mean age 37.0 years). Usability was rated high (CUQ mean 76.93, SD 12.51). Response quality and feedback quality received the highest ratings (Cronbach's α = 0.90 and 0.86, respectively), while realism scored lower (α = 0.87). Despite limited perceived naturalness, most students reported behaving professionally, respectfully, and empathically toward the VP. Educational value was rated highly, especially regarding professional competence and communication skills. Qualitative findings highlighted interaction realism and the integrated feedback mechanism as the most valued aspects, alongside requests for more unpredictable patient behavior and multimodal interaction formats. Conclusions AI-based virtual patients with integrated automated feedback are perceived as a valuable, scalable tool for anamnesis training across diverse health professions, extending prior evidence beyond medical education. While realism and emotional expressiveness remain areas for improvement, findings support the use of such systems as an effective formative learning tool. Future research should investigate the transfer of acquired skills to clinical practice and long-term learning outcomes.</p>