Breaking bad news with reflective AI virtual experience (BRAVE): development and evaluation of a generative AI-based simulation tool

BACKGROUND: Breaking bad news (BBN) to patients is complex and emotionally and cognitively demanding. The way bad news is delivered has implications for patients, their significant others, and healthcare professionals. Despite the development of structured BBN protocols, many health professionals report insufficient training and low self-efficacy for conducting these conversations. While best-practice BBN education exist, opportunities for repeated practice with high-quality feedback are limited, representing a critical barrier to scalability and global health equity, particularly in resource-constrained settings. We developed BRAVE (Breaking Bad News with AI Reflective Virtual Experience), a generative AI-based simulator that includes a written BBN encounter with a virtual patient. Unlike prior AI simulators that primarily offered structured summary feedback, BRAVE explicitly combines a prompted, dialogic reflective debrief followed by personalized feedback mapped to the SPwICES protocol. METHODS: A prospective, single arm, pre-post design, which incorporates both quantitative and qualitative measures. Ninety-nine senior medical students completed one BRAVE session following a required BBN course. Pre/post questionnaires assessed self-efficacy, willingness to engage in BBN, and intention to apply the BBN protocol. User experience was rated on a 5-point scale. Quantitative analysis included paired-sample t-tests and descriptive statistics. Qualitative data from open-ended responses were analyzed using immersion/crystallization. RESULTS: Self-efficacy increased from M = 6.95 to M = 7.62 (p