Do Artificial Intelligence Clients Speak Like Human Clients? Exploring GPT‐4's Content‐Level Performance in Counseling Role‐Play

Download PDF
Open Webpage

Xihe Tian, Yanhong Liu, Keni Li Zhang, Sarah Litt, Kahyen Shin, Tanqiu Jiang

Published: 05 Nov 2025, Last Modified: 19 Nov 2025Journal of Counseling & DevelopmentEveryoneRevisionsCC BY-SA 4.0
Abstract: Traditional role-playing methods in counselor education, such as peer counseling and actor-based simulations, enhance trainee competencies but have also created ethical dilemmas. These methods are not cost effective and may not capture and address client diversity. With the advancement of artificial intelligence (AI), new opportunities have emerged for enhancing training through AI-driven simulations. This study explores the use of GPT-4, a large language model (LLM) with voice capabilities, for client simulation in counseling role-play practices. We created 12 fictional client scenarios and engaged in simulation sessions with the fictional clients via GPT-4, then assessed its content-level performance using qualitative content analysis. Findings indicated that GPT-4 showed high-level performance in generating client narratives that include realistic descriptions of presenting concerns, emotional experiences, and culturally relevant descriptions. These results suggest that GPT-4 may offer a cost-effective, accessible, and useful training tool for counselors to independently practice and enhance skills.