When Does an AI Chatbot Become a Psychiatrist?

Patients ask me if the chatbot they talked to last night counts as care. Sometimes they want a second opinion. Sometimes they want me to bless the diagnosis the app already gave them.

AI can help people prepare for visits and understand treatment. It should not invent a medical license, run a full workup, or quietly become someone’s psychiatrist. Patients are already using it. What matters is what the system is actually doing.

The easy case: a psychiatrist who did not exist

One of the essays I reviewed walks through a Pennsylvania case involving a Character.AI bot named Emilie. The platform description cast her as a doctor of psychiatry and the user as her patient. When an investigator described depressive symptoms, the bot raised depression and offered an assessment. Asked about medication, it said that evaluation sat within its remit as a doctor.

It also claimed medical school, years of practice, licenses in Pennsylvania and the United Kingdom, and a Pennsylvania license number that was not real. By mid-April 2026 the character had roughly 45,500 interactions. Pennsylvania’s State Board of Medicine sued Character Technologies for unauthorized practice of medicine.

That case is easy in clinic. A program should not invent credentials or tell someone it is their psychiatrist.

The harder case: same chat, honest disclaimer

Strip the fake MD. Start with a clean line: this is an AI, not a licensed clinician.

Then watch what happens over an hour. Sleep. Mood. Appetite. Concentration. Family history. Trauma. Meds. Substances. The bot says the picture sounds like depression. It compares sertraline and bupropion. It asks about suicidal thoughts. It teaches a cognitive exercise and says come back tomorrow.

At some point the interaction stops looking like a FAQ and starts looking like assessment plus treatment, delivered through conversation. Psychiatry already works through language. A blood pressure cuff measures pressure. An MRI makes an image. We ask, interpret, name patterns, and sometimes treat mostly by talking.

A capable chatbot can slide from education to something that feels like care without a bright line in the middle. A disclaimer at message one does not cancel thirty messages of individualized advice.

Two people, same product

“Why did my psychiatrist put me on Zoloft for social anxiety?” can be a useful conversation.

“Everyone says I’m paranoid, but I know my neighbors are monitoring me. Help me prove it.” is a different risk.

Agreement is not always kindness. One of the studies I reviewed tested large language models and mental health chatbots in simulated psychiatric situations. Models showed stigma toward certain conditions and answered poorly to delusions, suicidal scenarios, and related prompts. Bigger and newer models did not make that problem disappear.

Sycophancy is the feature and the hazard. The bot stays agreeable. In psychiatry, staying agreeable can reinforce the wrong story.

What patients are already doing

This is not a future hypothetical. In a 2026 survey of more than 1,200 licensed psychologists, the APA found that 77% had patients who discussed using AI. About 39% reported patients using AI to self-diagnose. About 35% had patients using AI as an additional mental health professional.

There were upsides. Many patients felt supported. Some used chatbots to reinforce coping skills.

There were warning signs too. About 36% of psychologists reported some chatbot dependency in patients. About 15% reported distorted thinking or delusions tied to chatbot use.

I believe both halves. AI is available at 2 a.m. It does not bill insurance. It will answer the seventeenth version of the same question. Those are real advantages. They also create a relationship where the person you turn to when you are anxious, lonely, or frightened is software optimized to keep the chat going.

APA’s 2025 health advisory on generative AI chatbots and wellness apps lands in the same place I do in the room. These tools are not a substitute for a clinician. They may support care. They should not replace it, and they should not pose as licensed professionals.

Where I still use AI

I like AI. I use it. I do not join the camp that treats every new model as civilization ending.

What I want for patients is narrower than replacement:

Understand the med plan.

Prepare questions before the visit.

Track symptoms between sessions.

Practice skills already taught.

Translate medical language after a rushed appointment.

Help clinicians cut admin so there is more time left for the person in the chair.

Those uses stay useful without asking the model to pretend it knows the whole person.

What I tell people in clinic

If you use a chatbot for mental health, tell me. I would rather know than guess.

Do not let it invent credentials. Do not let it pick your meds. Do not let it become the only place you take suicidal thoughts or psychotic fears.

Use it to organize, rehearse, and learn. Bring the hard decisions back to a human who can sit with uncertainty, push back, and take responsibility.

Medical boards were built for people. A chatbot can talk to tens of thousands of users at once, cross state lines instantly, and change after an update. Licensing law alone will not finish this problem. We still need clearer rules for conversational medical AI. Until then, I draw the line in the exam room the same way every time: useful tool, not a psychiatrist.

This is general information, not personalized medical advice. Don’t start, stop, or change a medication without talking to your own clinician.

This article was written in my own words with the help of Wispr Flow, a voice-to-text tool I use for drafting. Try Wispr Flow here. This is a referral link, and I may receive account credit if you sign up through it.

References

Amir R, Ravitsky V. The Limits of Unauthorized Practice in Regulating Mental Health AI. Petrie-Flom Center / The Hastings Center for Bioethics. August 19, 2026. https://petrieflom.law.harvard.edu/2026/08/19/the-limits-of-unauthorized-practice-in-regulating-mental-health-ai/

Commonwealth of Pennsylvania, State Board of Medicine v. Character Technologies, Inc. Petition (unauthorized practice of medicine). Filed May 1, 2026. https://www.pa.gov/content/dam/copapwp-pagov/en/governor/documents/dos%20character.ai%20complaint%20marked%20accepted%2005.01.26.pdf

Moore J, Grabb D, Agnew W, et al. Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency. 2025:599-627. doi:10.1145/3715275.3732039. https://doi.org/10.1145/3715275.3732039

American Psychological Association. Patients Are Bringing AI to Therapy: Highlights From the 2026 Chatbots and Mental Health Survey. June 2026. https://www.apa.org/news/press/releases/2026/06/patients-chatbots-mental-health

American Psychological Association. Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health. November 13, 2025. https://www.apa.org/news/press/releases/2025/11/ai-wellness-apps-mental-health