Takeaway
AI can feel remarkably supportive to patients, but fluent and empathic responses are not a substitute for clinical judgment. Ask patients directly about their AI use, clarify its limitations, and show how the genuine connection with an accountable healthcare provider is more powerful than any code.
Lifelong learning in clinical excellence | September 11, 2026 | 2 min read
By Jennifer Katzenstein, PhD, Johns Hopkins Medicine
When an adolescent patient told me that an artificial intelligence (AI) chatbot was the only “person” who truly understood her anxiety, I was struck by both the appeal and the risk. AI is always available, responds without visible judgment, often validates you, and can sound remarkably empathic. Yet a fluent response isn’t the same as clinical judgment. Our goal as healthcare professionals shouldn’t be to shame patients for using AI for mental health support, but to help them use it with appropriate skepticism and strong guidance. Key conversations should include:
1. Ask about AI use directly.
Make AI use a routine part of the clinical interview. Ask which devices, apps, software programs patients use, what they discuss with AI, and whether the interactions affect their mood or behavior. A curious, nonjudgmental approach surrounding online presence makes disclosure more likely, especially when advice involves self-harm, eating, trauma, relationships, and/or medication.
2. Clarify what AI can and can’t do.
Explain that generative AI produces plausible language, not verified clinical judgment. It may invent facts, miss context, or reinforce a patient’s assumptions. It can help patients brainstorm coping strategies or prepare questions, but it shouldn’t diagnose conditions, recommend medication changes, or replace a qualified professional. Encourage patients to bring their findings into appointments for discussion.
3. Protect privacy and personal information.
A chatbot may feel private, but patients shouldn’t assume that the interaction is confidential or protected like a clinical conversation. Prompts may be stored or reviewed. Counsel patients not to enter personal information, and to review each platform’s privacy settings.
4. Watch for emotional dependence and reinforcement.
Because AI is always available and often agreeable, some patients may prefer it to human relationships. Watch for lost sleep, withdrawal, distress when access is interrupted, or an inability to hear other perspectives. Patients experiencing psychosis, mania, severe depression, disordered eating, or intense loneliness may be especially vulnerable to responses that validate distorted beliefs or unsafe behavior.
5. Create a human safety plan.
Be explicit that a chatbot isn’t an emergency service. Patients using AI for emotional support should know whom to contact when distress escalates. If AI suggests self-harm, violence, stopping treatment, or concealing symptoms, end the interaction and contact a trusted person or clinician. For immediate danger, call 911 or go to the nearest emergency department; patients can also access the Suicide Prevention Lifeline by calling or texting 988.
AI will become increasingly embedded in daily life. We must teach patients to question it, protect themselves, and recognize when human care is essential. AI may support reflection and access to general information, but it can’t replace the context, accountability, and genuine connection at the heart of care.
This piece expresses the views solely of the author. It does not necessarily represent the views of any organization, including Johns Hopkins Medicine.