AI Large Language Models Use Cases, Barriers, and Ethical Tensions across Stakeholders

Introduction: Social workers routinely participate in problem-solving processes that require increased knowledge, substantial communication, documentation, and interpretation of policies. Artificial Intelligence (AI) large language models (LLM) present new opportunities to support these tasks through its function of text generation. Simultaneously, social work practice is guided by values, ethics, and federal and state-level regulations designed to ensure client safety and confidentiality. This qualitative study explores use cases in social work and identifies factors that facilitate or discourage their adoption among key stakeholders within a social work school at a public university, including students, faculty and staff, and practicum supervisors.

Methods: Six focus groups were conducted with students (n = 9), faculty and staff (n = 11), and practicum supervisors (n = 8) affiliated with a BSW and MSW program at a large research-intensive university in the Southwest United States. Each group participated in one of two 90-minute Zoom sessions. To protect confidentiality, participants used pseudonyms and kept cameras off. Semi-structured questions explored experiences with AI LLM use in practice, perceived barriers and facilitators, guidance from colleagues, mentors, and professors, and any ethical concerns as to using AI LLM in practice. Participants represented diverse roles and settings: 44% and 56% of students were enrolled in BSW and MSW programs, respectively; faculty and staff included tenure-track (9%), career-track (64%), and administrative staff (27%); and practicum supervisors primarily worked in behavioral health (50%), education (38%), and policy setting (10%). Regarding the frequency of AI LLM use for work-related tasks, 33% of students reported using them at least a few times per week, compared to 36% of faculty and staff and 50% of practicum supervisors. Applied thematic analysis was used to identify patterns in use cases and factors influencing adoption.

Results: Participants identified three primary functions of AI LLM in social work practice: (1) documentation support, (2) communication support, and (3) expanded access to information. As to most frequently reported use cases, documentation support included clinical note writing, improving professional tone, outlining documents, and editing grammar. Communication support involved language translation, clarifying complex concepts, and drafting emails or psychoeducational materials. Expanded access to information included assistance with tasks related to data management, literature searches and organizing information, and simplifying complex policy documents. Despite these benefits, participants expressed significant concerns, including conflicts between personal values and AI LLM adoption, confidentiality and data security risks, potential inaccuracies (e.g., AI hallucinations), and uncertainty about AI LLM allowed in clinical contexts. A lack of formal training emerged as a major barrier to adoption.

Discussion and Conclusions: Findings highlight that while AI LLM are intuitive and accessible, their adoption in social work is shaped less by technological barriers and more by ethical concerns and value alignment. Unlike prior innovations, AI raises fundamental questions about professional responsibility, confidentiality, and trust. Effective AI integration in social work may be less a technical challenge than a professional development and governance challenge. Investments in training, policy infrastructure, and cross-sector dialogue between educators, agencies, and professional associations are the most actionable next steps.