Fostering reflection: Development and initial evaluation of a mentalization-based GenAI simulator for foster care supervisors

Background: Foster care supervisors require advanced reflective functioning (RF) to manage complex relational demands, yet traditional training is difficult to scale and access. Generative AI (GenAI) may help by enabling simulated practice environments. Objective: This feasibility study describes the development and evaluates the initial usability and acceptability of a GenAI-based chatbot simulator designed to support RF among foster care supervisors. Participants and setting: Forty-six female foster care supervisors (mean age = 41.6 ± 8.8 years; mean field experience = 8.9 years) participated. Methods: Using a mixed-methods design, participants completed a single session with a GenAI simulator featuring an interactive dialogue with a virtual foster mother and a multi-layered feedback system. Post-session user experience was assessed via quantitative ratings and open-ended questions; quantitative results were summarized descriptively and explored with one-sample tests against the scale midpoint, and qualitative responses were analyzed using thematic analysis. Results: Usability and acceptability were high. Feedback was rated as highly contributory (5.83 ± 1.40 on a 7-point scale), and willingness to recommend the tool was strong (6.67 ± 0.77). Qualitative findings suggested the simulator functioned as a safe space for reflection, alongside a dialectic between valuing an authentic, supportive interaction and wanting higher-friction exchanges that more closely resemble clinical resistance. Conclusions: This GenAI-based simulator appears highly acceptable as an RF training tool for foster care supervisors and may complement traditional training by providing a scalable, accessible, and safe practice environment.