AI Creates the Patient and the Student Does the Reasoning
Featured faculty: Tara Mansour
Associate Program Director
Doctor of Medical Sciences Program
Bouvé College of Health Sciences
TL;DR: A chatbot acting as various patients to help students think through their clinical reasoning.
What she’s doing: Tara built an AI-based chatbot practice tool for a single occupational therapy student who was stuck on the assessment section of a SOAP note –a standard format for documenting a clinical encounter, made up of subjective, objective, assessment, and plan sections– the part of clinical documentation where a student must move from reporting what happened to explaining what it means. This project predates her arrival at Northeastern, developed while she was the Academic Fieldwork Coordinator, or Director of Clinical Education, in an entry-level OT program, but it shaped how she now thinks about customizing student support to the specific gap a learner is facing rather than applying a generic fix. Tara front-loaded the tool with de-identified exemplars, the structure of a SOAP note, and strict role instructions: the AI would generate a fictional patient profile and the subjective/objective sections, but the student had to write the assessment and plan on her own before receiving any feedback. The student’s only input was a fictional name and age. Everything else, diagnosis, context, the everyday activities disrupted by injury, was generated to support repeated, low-stakes practice without touching real patient data. Of significance, the tool was never designed to replace the expertise, judgment, or feedback of the fieldwork educator. It was created as a supplemental practice tool, a way for the student to rehearse the reasoning process repeatedly between opportunities for direct teaching and feedback from the clinician supervising her learning.
What’s working and what isn’t: The student reported improvement, and her fieldwork educator’s formal evaluation reflected the same growth. In this process, Tara discovered a more consequential concern that wasn’t about the AI’s design at all. Despite encouragement, the student never told her fieldwork educator she’d used the tool because she worried about how disclosure might be received. That gap, Tara argues, is the real risk: a fieldwork educator’s site-specific expertise could have sharpened the tool further, but transparency anxiety kept that expertise out of the loop entirely. The lesson Tara draws is clear: blanket prohibition doesn’t stop AI use, it just pushes it underground.
| Adapting Across Contexts: A PT program could do the same for a plan-of-care narrative, or a speech-language pathology program for a diagnostic impressions section, in each case letting AI supply the case material while reserving the reasoning step for the student. Outside healthcare, a law school could apply the same structure to the analysis section of a legal memo, having AI generate the fact pattern and relevant statutes while the student writes the argument connecting them unassisted. Discussions about transparency are still needed: if a program wants students disclosing AI use to supervisors, that expectation has to be stated explicitly and modeled by faculty, not assumed. What is the equivalent “assessment section” in your field, the step where students must show reasoning rather than recall? What fictional or de-identified case material could stand in for real records? And what would need to be said out loud so students don’t treat disclosure as a risk? |
What’s next: Tara wants to move past perception-based research since most existing studies ask whether students feel AI tools helped rather than measuring whether documentation skills actually improved. She is hoping to begin additional research soon. Longer term, she’s aiming further upstream: developing AI literacy competencies for health professions educators across OT, PT, speech-language pathology, and nursing, so faculty can design these activities deliberately rather than reactively.
Have you tried something new with AI? How did it go? Send an email to [email protected] to let us know!