Cases and Lessons in Prompting Artificial Intelligence for Social Work

Since the public release of ChatGPT in late 2022, Large Language Models (LLMs) as a form of Generative Artificial Intelligence (AI) have emerged as a transformative technology with the potential to create value across diverse domains. A central new practice is prompting: articulating requirements in the form of instructions to conversational agents via natural language. The challenges faced by innovators, especially non-technical ones, in adopting these new tools are unknown. Social workers, professionals who serve society’s most vulnerable members, have historically been reluctant to embrace new technologies. This paper presents examples and lessons learned from introducing social work students to LLMs, and prompting specifically, via the development of custom conversational agents. Participants (n = 28) created CustomGPTs tailored to their professional use cases, engaging in a guided process to design and refine the prompts. We describe our approach, report the cases, and examine challenges. These insights can inform the training of future innovators, the evolution of conversational agent platforms to better support the creation of robust, trustworthy AI tools by non-technical users, and serve as empirical data for innovator involvement in the requirements engineering process.