Can an artificial intelligence (AI) chatbot serve as a reliable tool for mental health clinicians? A new study in Nature’s Scientific Reports shows how an AI assistant based on OpenAI’s Large Language Model (LLM) GPT-4 architecture achieves high accuracy in performing clinical patient diagnostic interviews for common mental health disorders.
“Our study demonstrates the promise of AI-powered clinical interviews, powered by LLMs, as a reliable and scalable innovation in mental health diagnostics,” wrote lead author and psychology professor Sverker Sikström at Lund University in Sweden, along with co-authors Rebecca Astrid Boehme, Mariam Mirström, Thibaud Agbotsoka, Gergő Győri, Marta Lasota, Mona Tabesh, Lotta Stille, and Danilo Garcia.
The use of AI in mental health is rising. In one recent poll conducted by the American Psychological Association’s 2025 Practitioner Pulse Survey, over half of the 1,742 psychologists polled, 56% of participants, reported using AI to help with their work at least once in the past year, a huge increase compared to 29% the year prior. Among participants using AI to help with their work as psychologists, the most common purpose was for routine tasks such as drafting email and other content, and summarizing clinical notes or articles, and note taking. AI-assisted clinical interviews may be next among the most prevalent uses in the near future.
“By evaluating the validity of AI-powered clinical interviews and their capacity to provide a positive, person-centered user experience, this study offers evidence that such tools may help address key limitations in current diagnostic practices, such as, scalability, standardization, and accessibility,” the researchers wrote.
The researcher recruited 550 participants for this study, with 100 healthy controls. The remaining 450 participants have self-reported conditions that were clinically diagnosed by professionals. Out of those diagnosed with mental health disorders, they were further subdivided into diagnoses groups of 50 participants each for post-traumatic stress disorder (PTSD), attention-deficit hyperactivity disorder (ADHD/ADD), generalized anxiety disorder (GAD), major depressive disorder (MDD), obsessive-compulsive disorder (OCD), bipolar disorder (BD), eating disorders (ED), autism spectrum disorder (ASD), and substance use disorders (SUD). This pool was further filtered for data quality, which narrowed it down to 303 participants consisting of 55 controls and 248 with self-reported mental health conditions that were professionally diagnosed.
The study used an online AI assistant called TalkToAlba, a platform currently used by Swedish and European clinicians. Lead author Professor Sikström is the founder of Talk To Alba. The AI assistant was developed on OpenAI’s GPT-4, specifically GPT-4 Turbo Preview, to estimate if a participant had a 50% or greater probability in meeting the DSM-5 diagnostic criteria for one of the nine mental health disorders selected for this study. DSM-5, short for Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, is a professional reference book and guide for brain-related conditions and mental health by the American Psychiatric Association.
The multimodal AI tool based its predictions on responses collected from participants containing answers to 15-20 open-ended mental health questions, either via text or speech. After the interview, the participants in turn assessed the AI assistant for understanding, relevance, supportiveness, and empathy, described their user experience with five adjectives, and compared it to standard ratings and clinical interviews.
“The AI system achieved diagnostic accuracy that was comparable, and in some cases superior, to widely used rating scales across several common psychiatric disorders,” reported the researchers.
Interestingly, the most-reported participant descriptive words for the AI assistant were helpful, caring, understanding, interesting, and informative. Further, the participants rated the AI assistant as highly supportive, empathic, understanding, and relevant.
“Notably, our results demonstrate that diagnostic precision can be achieved without sacrificing user experience, suggesting AI-powered clinical interviews can complement traditional clinical workflows while also offering additional benefits, including accessibility, standardization, low-cost alternatives, and enhanced patient experiences,” the team reported.
This pioneering study signals the potential of AI in mental health care. Mental health professionals are adopting AI-assisted tools to automate or assist with routine tasks. Whether or not AI-assisted clinical interviews will become a common practice among psychologists and other mental health professionals in the future is unfolding now in the present.
Copyright © 2025 Cami Rosso All rights reserved.