Simulating the Patient: A Case in Building AI Chatbots for Classroom Practice

What does it take to build an AI chatbot that more reliably supports student practice, and what breaks along the way? This session presents a case of faculty-technologist collaboration to create an AI chatbot for an undergraduate psychopathology course. Students used the chatbot to practice cognitive behavioral therapy techniques, particularly cognitive restructuring, in a low-stakes environment. The bot provided limited feedback and prompted the students to return their transcripts to the instructor for evaluation, keeping content expertise and assessment with the faculty member where they belong. This session is more about the craft of system prompting as it is about the pedagogical design. Writing a prompt that behaves more consistently over many unpredictable interactions and various patient background profiles turned out to be difficult, and remains a skill even as AI models improve. One challenge: using AI to help draft a system prompt is helpful, but can also introduce accidental contradictions disrupting instructions. It takes human review to catch what changed. Attendees will gain a picture of what this kind of project involves: scoping the simulation, iterating on prompts, and building collaboration with others in the design process. Whether you are exploring AI classroom integration for the first time or already supporting active projects, this session offers insights on lessons learned from a challenging build.