![[ Visit Client Website ]](images/banner.gif)
These studies also demonstrate that AI holds significant promise for improving behavioral health. LLMs can enhance detection of stigmatizing clinical language shaping referral patterns and outcomes (Study Five) and accelerate construction of measurement tools for person-centered care in documentation (Study Three). These technical capabilities must be matched by training and policy action. Collectively, the studies call for AI literacy across the profession, regulatory frameworks balancing innovation with protection, documentation standards that reduce bias, and inclusive approaches centering those most affected. Advancing AI in social work is not only a technical challenge but a professional and ethical imperative.
Study One: This study analyzed how Utah, Nevada, and Illinois: the first states to legislate AI mental health chatbots, framed regulatory problems differently, producing distinct approaches: consumer protection, licensing-based prohibition, and credentialing extensions respectively. None addressed general-purpose LLMs used therapeutically. The findings highlight how problem construction shapes whose interests are protected and which risks remain unaddressed.
Study Two: This qualitative study interviewed 38 AANHPI young adults about using AI chatbots for mental health support. Participants valued AI's anonymity to bypass cultural stigma, often turning to it after negative therapy experiences. Higher mental health literacy enabled increased use. Findings suggest AI serves as a supplement rather than a gateway to formal care.
Study Three: This study used six LLMs to generate potential terms for a person-centered care dictionary for behavioral health documentation. Expert annotators reviewed 432 LLM-generated words, achieving agreement (AC1 = 0.906) and retaining 408 terms. The method offers a replicable, cost-efficient alternative to traditional dictionary construction for large-scale clinical text analysis.
Study Four: This qualitative study found that social workers using AI documentation tools develop informal "folk theories" about how these systems handle data, due to limited AI literacy and absent professional guidance. These theories shape clinical behaviors around consent, privacy, and recording practices, highlighting urgent needs for AI literacy training and regulatory frameworks.
Study Five: This study uses natural language programming enhanced by large language models to detect stigmatizing language in clinical notes for patients with substance use disorders. Unprofessional language appeared in 69.1% of SUD notes and was independently associated with higher behavioral health referral odds, suggesting documentation bias may drive referral patterns rather than clinical need alone. The authors recommend stigma-free documentation training and EHR-integrated language alerts.