Structured human-LLM interaction design reveals exploration and exploitation dynamics in higher education content generation

Large Language Models (LLMs) present a radically new paradigm for the study of information foraging behavior. We study how LLM technology is used for pedagogical content creation by a sample of 25 participants in a doctoral-level Artificial Intelligence (AI) in Education course, and the role of computational-thinking skills in shaping their foraging behavior. We used editable prompt templates and socially-sourced keywords to structure their prompt-crafting process. This design influenced participants' behaviors towards exploration (to generate novel information landscapes) and exploitation (to dive into specific content). Findings suggest that exploration facilitates navigation of semantically diverse information, especially when influenced by social cues. In contrast, exploitation narrows the focus to using AI-generated content. Participants also completed a Computational Thinking survey: exploratory analyses suggest that trait cooperativity encourages exploitation of AI content, while trait critical thinking moderates reliance on participants' own interests. We discuss implications for future use of LLM-driven educational tools.