How to Counsel Patients About Generative AI Mental Health Chatbots
How should clinicians counsel patients who want to use generative AI chatbots for mental health support?
Patients are already using generative AI chatbots for emotional support, coping help, and mental health advice, often outside clinical oversight. In primary care and psychiatry, the immediate task is to determine what the patient wants from the tool, whether it is being used safely, and whether it is supplementing or replacing professional care.
-
Assess what the patient wants the tool to do
Start by asking what need the patient is trying to meet and match that need to the type of generative AI mental health tool being considered. The article specifically recommends clarifying what service the platform provides, such as cognitive-behavioral therapy, motivational interviewing, monitoring, emotional support, symptom tracking, or advice, and how interactive and personalized the tool is.
-
Determine whether the tool is being used as an adjunct or a substitute
Clarify whether the patient is using the chatbot to support self-management between visits or instead to replace human care and human connection. The article advises recommending generative AI tools only as adjuncts to care and avoiding their use as substitutes for professional treatment, especially when clinician assessment or crisis management is required.
-
Screen for high-risk situations where chatbot use is unsuitable
Identify whether the patient has active thoughts of suicide, psychosis, severe substance use, or other situations involving significant risk of harm. The article explicitly states that recommendations for generative AI tools as a substitute for care should be avoided in high-risk states and notes that chatbots may be unable to manage crises or appropriately escalate acute psychiatric symptoms.
-
Review the evidence for the specific intended use
Discuss whether there is scientific evidence supporting the tool's proposed role, such as symptom reduction, self-management, or adjunctive monitoring. The article notes short-term benefits for anxiety and depression in some studies, with the most robust benefits after 8 weeks, but also emphasizes that long-term efficacy is uncertain and that many trials compared chatbots with waitlist, information-only, or other nonexpert interventions rather than clinician-delivered psychotherapy.
-
Explain the tool's practical benefits and limitations
Review both advantages and disadvantages in concrete terms. The article notes benefits such as 24/7 availability, anonymity, and scalable behavioral techniques, but also warns that these tools may fail to recognize nuance, make factual errors, mirror rather than challenge distorted thoughts, and be unable to manage crises.
-
Counsel about privacy and data governance
Review where and how the app stores data, whether data are encrypted, whether personal data may be used to train models, and which legal jurisdictions oversee data processing. The article recommends steering patients toward tools with transparent privacy policies, clear consent processes, secure data handling such as encryption or pseudonymization, and options to delete their data, while recognizing that many direct-to-consumer tools are outside HIPAA.
-
Advise patients how to use chatbots safely
Give concrete examples of appropriate and inappropriate use. The article recommends encouraging patients to use generative AI tools for self-management, symptom monitoring, and homework or skills practice between sessions, and to avoid relying on an unmonitored chatbot during crises such as suicide-related distress.
-
Document the discussion and guide tool selection
Document the counseling conversation in the medical record, including risks discussed and the intended scope of use. When possible, the article recommends steering patients toward programs that comply with local standards or have been independently evaluated without industry biases, and advising patients to look for peer-reviewed evidence, update frequency, and available customer support.
Clinical Considerations
- Evidence for chatbot benefit is strongest for short-term symptom reduction and did not show substantial persistence at 3-month follow-up in one meta-analysis.
- Many chatbot trials compared generative AI with waitlist, information-only, or psychoeducation controls rather than expert clinician-led therapy.
- Direct-to-consumer tools are generally unregulated and may have fewer safeguards for privacy, accuracy, and clinical validity.
- Algorithmic bias, demographic performance disparities, and stigmatizing responses remain clinically important concerns that require further evaluation.
Bottom Line
Counsel patients to use generative AI mental health chatbots only as adjuncts for low-risk self-management tasks, not as substitutes for professional care or crisis evaluation.
