Name
Enhancing Standardized Patient Script Accuracy Through AI-Assisted Review Using ChatGPT
Authors

Adriana Bautista, Florida International University Herbert Wertheim College of Medicine
Javier Fernandez, Florida International University Herbert Wertheim College of Medicine
Maria Stevens, Florida International University Herbert Wertheim College of Medicine

Date & Time
Thursday, October 23, 2025, 12:45 PM - 12:59 PM
Presentation Category
AI & Technology
Description

Purpose
Ensuring accuracy and consistency in standardized patient (SP) scripts is essential for delivering high-quality simulation-based education. The FIU-HWCOM SP Program explored a novel approach using generative AI (ChatGPT) to assist in reviewing SP scripts for context inconsistencies.

Methods
Scripts followed a standardized template, enabling development of a consistent ChatGPT prompt to guide the AI through key content areas. ChatGPT was tasked with identifying inconsistencies across four predefined categories: Timeline (e.g., mismatched dates or ages), Symptoms (e.g., conflicting complaints), Communication and behavioral cues (e.g., unclear tone or nonverbal behaviors), and Medical content alignment (e.g., deviations from faculty-approved details). The AI-generated results were analyzed based on the prevalence (how many scripts were affected) and depth (number of inconsistencies per script). Team members independently reviewed all flagged items and recorded agreement rates for each category.

Results
ChatGPT detected at least one inconsistency in all 33 SP scripts across the four categories. Most scripts contained multiple inconsistencies. Reviewers agreed with 72.7% (24 scripts) of timeline flags, 60.7% (20 scripts) of symptom-related issues, 66.7% (22 scripts) of communication/behavioral, and 81.9% (27 scripts) of medical content discrepancies. These results suggest that ChatGPT can effectively support SP script quality assurance when used in combination with human review.

Conclusion
This study highlights ChatGPT’s potential as a scalable tool to assist simulation teams in identifying script inconsistencies. However, effective integration of AI into health education requires ongoing evaluation and thoughtful implementation. Future steps include assessing AI-driven improvements in SP performance and continuing human-AI result comparisons. As AI advances, sustained dialogue will be key to ensuring its ethical and effective use in medical education.