The mismatch between demand for behavioral health care and available clinicians is severe enough that conversational tools are being deployed as an interim measure by payers, employers and health systems alike. The evidence supporting that deployment is narrower than the deployment.

Published trials generally recruit adults with mild to moderate symptoms, exclude active suicidality, run for six to twelve weeks, and measure symptom scales. Results in that population are modestly positive and reasonably consistent. The population being served by real deployments includes people the trials excluded.

Crisis handling is where the concern concentrates. Most tools route explicit crisis language to human support, and evaluations of that routing find it works for explicit language and much less well for indirect expression, which is the more common presentation.

Several clinicians we spoke with made a point that is easy to lose: the comparator is not a therapist. It is a waitlist. That framing justifies deploying an imperfect tool and does not justify skipping the study, and at present the field is doing the first without the second.