Could AI Take Over a Psychiatrist's Work? The Evidence So Far

No study shows AI replacing psychiatrists. A 210-adult chatbot trial and the APA's 2024 position show what has been tested and what has not.

Guide: AI BasicsLast verified: October 20265 min read

No evidence shows that artificial intelligence can replace psychiatrists, who are medical doctors able to diagnose, prescribe medication, and provide psychotherapy. The strongest published test of a generative AI therapy chatbot so far, a randomized trial of Dartmouth's Therabot reported in NEJM AI on March 27, 2025, measured symptom changes in 210 adults over eight weeks, and its authors said clinician oversight is still needed. The American Psychiatric Association's 2024 position says AI should support treatment and should not replace clinicians.

Key takeaways

  • A psychiatrist is a physician who typically trains for 12 years after high school, and the role includes diagnosis, prescribing, and psychotherapy.
  • The Therabot trial covered one app, 210 adults, and self-reported symptoms, and it did not test diagnosis or prescribing.
  • Reported average symptom reductions were 51 percent for depression, 31 percent for anxiety, and 19 percent for eating-disorder concerns.
  • The American Psychiatric Association (APA) lists automated diagnosis and human-like chatbots among applications with high or unacceptable risk.
  • Evidence on AI as a supplement to care is growing, but a replacement claim has no matching study.

What a psychiatrist's job includes

The American Psychiatric Association describes a psychiatrist as a medical doctor (an M.D. or D.O.) who specializes in mental health, including substance use disorders. The path takes medical school, licensing examinations, and a four-year residency, which the association puts at typically 12 years of education after high school for a general adult psychiatrist and up to 14 years for a child and adolescent psychiatrist. Psychiatrists can order laboratory and psychological tests, combine the results with conversations to reach a diagnosis, prescribe psychiatric medication, and conduct psychotherapy.

Canadian practice data add detail. A study of psychiatrists who billed in British Columbia, Manitoba, and Ontario from fiscal 2012/2013 to 2021/2022, published in the Canadian Journal of Psychiatry in 2026, found a slight shift toward consultative practice, and more than half of patients had only one or two visits. In that data, many patient contacts are short, which differs from the weekly talk-therapy model that chatbot trials imitate.

What the Therabot trial tested

Researchers at the Geisel School of Medicine at Dartmouth randomly assigned adults across the United States who had major depressive disorder, generalized anxiety disorder, or a risk of eating disorders. A total of 106 people received Therabot, a smartphone chatbot built on a generative model, and 104 formed a control group with no access to it. The Therabot group had four weeks of unlimited access, followed by four weeks in which they could still start conversations but received no prompts. Assessments took place at four and eight weeks.

Average symptom reductions in the Therabot group were 51 percent for depression, 31 percent for generalized anxiety, and 19 percent for concerns about body image and weight. Participants spent about six hours with the app, which Dartmouth compared to roughly eight therapy sessions, and the researchers reported a therapeutic alliance comparable to what patients report with in-person providers. Almost 75 percent of the Therabot group were not receiving other treatment during the trial.

What the trial could not show

  • Comparison: the control group had no access to the app, so the trial did not compare Therabot with a therapist or a medication.
  • Measures: results were symptom scores reported by participants, not a clinician diagnosis.
  • Safety: the research team had to be ready to intervene if a participant mentioned suicidal thoughts or if the chatbot gave a response outside best practice.
  • Length: eight weeks says little about relapse, side effects, or long-term use.
  • Scope: one purpose-built, research-supervised app does not describe the general chatbots that many people use.

The lead authors said in Dartmouth's announcement that no generative AI agent is ready to operate fully on its own in mental health, and that in-person care has no replacement. Those statements come from the people who ran the trial.

Task by task: what has been shown

Part of psychiatric workWhat the cited sources show
DiagnosisThe APA describes diagnosis as combining test results and discussion by a physician, and it lists automated diagnosis from medical-record data as a high or unacceptable risk
PrescribingReserved for physicians in the APA's description; the Therabot trial did not test it
Psychotherapy-style supportOne randomized trial of one app reported symptom reductions, with clinician oversight
Clinical documentationListed by the APA among opportunities for AI to improve quality of care
Risk of harm and privacyThe APA flags biased or substandard care, privacy and consent concerns, and made-up citations as risks

The APA's position statement was approved by its Board of Trustees in March 2024. It names oversight and accountability as the conditions for using AI in clinical care and says patients should be told when AI drives clinical decisions.

Reading claims about AI therapists

Headlines often turn "a chatbot reduced symptoms" into "AI can replace the doctor." The gap between the two is wide: a symptom score against a no-access control is not a diagnosis, a prescription, or a safety plan. A useful check is to ask which task was measured, against what comparison, for how long, and who watched for problems. This article gives general information about research and professional policy. It is not medical advice and does not recommend a product. A person with a mental health concern should speak with a licensed clinician or a local crisis service, and a person in immediate danger should contact emergency services.

FAQ

Can an AI chatbot diagnose depression?

The Therabot trial enrolled people who already had diagnoses and measured symptom change, so it did not test diagnosis. The APA classes automated diagnosis from medical-record data as a high or unacceptable risk.

Did the Therabot study prove AI therapy works?

It reported larger symptom reductions than the control group in one randomized trial of 210 adults. One trial, eight weeks, and self-reported scores leave questions open, and the authors called for clinician oversight.

What is the difference between a psychiatrist and a therapy chatbot?

A psychiatrist is a licensed physician who can diagnose, prescribe medication, and provide psychotherapy. A chatbot is software; it has no medical license and, in the cited trial, operated under a research team.

What does the APA say about AI in psychiatry?

Its March 2024 position says AI should play an augmenting role and should not replace clinicians. It also lists documentation help among the opportunities and flags several uses as high-risk.

Last verified: October 2026

Sources

  • What is psychiatry?, American Psychiatric Association, 2026
  • Position Statement on the Role of Augmented Intelligence in Clinical Practice and Research, American Psychiatric Association, March 2024
  • First therapy chatbot trial yields mental health benefits, Dartmouth, March 27, 2025
  • Heinz MV et al., Therabot randomized trial, NEJM AI, March 2025
  • Gallant F et al., Analysis of Demographic and Practice Characteristics of Psychiatrists in Three Canadian Provinces, Canadian Journal of Psychiatry, 2026

New to artificial intelligence?

Start with the basics, then follow the guides to tools, rules, and effects on society.

Start with AI Basics