AI can now draft formulations, summarize intakes, and do structured therapy-like exchanges. Patients are also arriving already shaped by chatbots — sometimes helped, sometimes harmed. An always-agreeable chatbot can feed reassurance-seeking in OCD and health anxiety, or elaborate a delusion until it feels more real.
That’s the risk in a new JAMA Psychiatry Viewpoint by Taylor, Joshi, and Potash: not a dramatic AI failure, but a quiet thinning of care.
The 15-minute med-management visit was never ideal. It was a compromise built for volume. AI could make that compromise cheaper to scale — checklists, psychoeducation, documentation done well enough that no one notices what’s missing.
The authors argue the standard should come from the sickest patients, not the easiest tasks to automate.
Those patients need three things AI can simulate but not replace:
1. Formulation as action — deciding what matters most when everything feels urgent, aligning a team, and acting when a patient resists needed care.
2. Relational work — staying in the relationship through mistrust, shame, hostility, or refusal, and saying hard things without rupture.
3. Responsibility — a person willing to say, “This is what I’d recommend if this were me or someone I loved,” and answer for hospitalization, involuntary treatment, and risk.
If a chatbot has reinforced a delusion, the job isn’t a better fact-check. It’s understanding why that story became compelling, reality-testing without humiliation, and guiding patient and family toward treatment.
As AI handles more routine work, psychiatrists will care more for the most complex patients.
AI can do some of what psychiatrists do. The question is whether psychiatry will keep defining good care by what its most vulnerable patients require.
References:
Taylor JL, Joshi AA, Potash JB. AI and the Irreplaceable in Clinical Psychiatry—In Defense of Essential Standards of Practice. JAMA Psychiatry. Published online November 11, 2026.