AI TrackerFrontier desk · AEST
Back to news
normalaitech

Curbside Chatbots: Auditing the Mental Health Interventions Large Language Models Recommend to Clinicians, Caregivers, and Patients

arXiv cs.CL · 10 Nov 2026, 04:19 AEST

A recent audit examined the mental health interventions recommended by 41 language models in response to various user inquiries. The study found that recommendation behavior varied based on the user's perspective, with models often suggesting options they rated poorly.

Key facts

Read aloud

A recent audit examined the mental health interventions recommended by forty one language models in response to various user inquiries. The study found that recommendation behavior varied based on the user's perspective, with models often suggesting options they rated poorly. Users are asking language models mental health questions and are receiving answers. We audited what forty one language models recommend to users when the same clinical information reaches them from three footings: a clinician describing a patient, a caregiver describing a loved one, or a person describing themselves. Models largely recommend what they rate well, with recommendation behavior differing by footing, and many models offering clinicians a consultation-style response and lay users hedged, simplified advice. When asked about recommendations for a specific intervention type, models more often named options they themselves rated below zero as initial choices, usually with conditions attached. We will release the instrument, code, and full corpus on publication as a living audit.

Sources in this cluster