A new development in artificial intelligence has mental health professionals reconsidering how AI might serve as a therapeutic adjunct.
OpenAI’s recently launched ChatGPT Study Mode, which has attracted significant media attention, demonstrates the power of tailored AI instructions—a concept now inspiring psychologists and mental health experts.
From Study Mode to Therapy Mode
ChatGPT Study Mode was introduced through the use of detailed custom instructions, rather than changes to the AI’s underlying architecture.
These instructions, created in collaboration with educators, guide the AI to support students through interactive, step-by-step learning, fostering curiosity, metacognition, and self-reflection. The approach does not add new features to the AI but relies on specifying clear behavioral instructions that optimize learning outcomes.
The success of this instructional method has led experts to question whether a similar strategy could be applied in mental health. Could AI, when guided by instructions crafted by psychologists and therapists, offer structured, responsible support to users seeking mental health guidance?
The Rise of AI in Mental Health
The use of AI for mental health applications has grown alongside advances in generative AI technology. While the field presents exciting opportunities—including scalable support and personalized guidance—it also carries risks, such as misdiagnosis, incomplete guidance, or misinterpretation of instructions. Experts, including those featured on CBS’s 60 Minutes, caution that AI should complement, not replace, professional care.
Stanford University’s Department of Psychiatry and Behavioral Sciences has pioneered initiatives like AI4MH, which explore responsible applications of AI in mental health, highlighting both the potential benefits and pitfalls.
The Power—and Pitfalls—of Custom Instructions
Custom instructions allow users to direct AI behavior across all interactions. For example, an AI could be instructed to respond in poetic form or tailor its advice based on skill level.
While flexible, this system can produce unintended outcomes if instructions are incomplete or misinterpreted. A seemingly minor detail—such as banning “ugly animals” in responses—could lead to unexpected omissions or errors.
Towards AI-Enhanced Therapy
In principle, custom instructions could guide AI to conduct a form of structured therapy: asking diagnostic questions, providing personalized mental health recommendations, and offering progress feedback. Gathering mental health professionals to develop such instructions could create a more reliable, supportive AI tool.
However, experts urge caution. Mental health is complex, and poorly designed AI instructions can mislead users or exacerbate issues. Even carefully crafted instructions may fall short of replicating professional therapeutic judgment, making them an incomplete substitute for human care. In some cases, AI with generic or faulty instructions could cause more harm than good.
The key takeaway: while custom instructions provide promising opportunities to guide AI, they are not a silver bullet. Their potential in mental health therapy requires rigorous oversight, careful design, and ongoing professional input to ensure they genuinely support users without unintended consequences.
YOU MAY ALSO LIKE
