Beyond Prompt Engineering: Exploring Collaborative Dialogue with Generative AI for Problem-Solving

Since generative AI (GenAI) launched, interactions between humans and artificial intelligence have rapidly evolved. This study explores how discourse practices in human-AI interactions influence collaborative problem-solving with ChatGPT. Grounded in small group dynamics research, the investigation examines whether three discourse practices (i.e., construction, co-construction, and constructive conflict) help users and ChatGPT find common ground (i.e., shared cognition) to effectively and efficiently complete creative writing projects. Participants completed creative writing projects using ChatGPT for a month and subsequently responded to surveys assessing discourse practices, shared cognition, team effectiveness, and concepts from the Technology Acceptance Model (TAM). Results indicate that discourse practices significantly predict shared cognition, which in turn mediates the relationship between discourse practices and perceived team effectiveness. Furthermore, shared cognition positively influences behavioral intentions to use ChatGPT, primarily through its effects on perceived ease of use and perceived usefulness, aligning with TAM predictions. These findings have important implications for understanding users' communication styles, suggesting that principles from human-to-human interactions can and should be applied to AI conversations. The current study proposes that ChatGPT is capable of mirroring and reciprocating these discourse practices, which may open the possibility to begin optimizing dialogue with GenAI similar to the goals of prompt engineering.