Abstract

Research on Generative Artificial Intelligence (GenAI) in education has concentrated on user attitudes and adoption intentions but less is known about how groups actually interact with such tools while solving complex problems. The present study treats prompting as observable, time-resolved, behaviour and investigates how a group's prompting develops and relates to task effectiveness over a collaborative session. Secondary (N = 114) and Higher (N = 57) Education students, working in self-formed groups (N = 45), completed a competitive mystery-solving task with unrestricted access to GenAI tools of their choice. Logged group prompts were coded with a scheme adapted from Bloom's Digital Taxonomy that distinguishes lower-order (Exploration and Action) from higher-order (Creation and Metacognition) cognitive behaviours. Groups produced a modest majority of higher-order prompts and shifted progressively from exploratory toward metacognitive prompting as sessions advanced. A higher proportion of higher-order prompting was associated with greater effectiveness which held even when the total prompt volume was controlled. Prior experience, general attitudes toward AI, and self-reported verification behaviour showed no reliable association with effectiveness. Perceived usefulness and prompting self-efficacy co-varied with success but are interpreted as concurrent correlates. Foregrounding what groups did with the tool and how that behaviour developed over time offers a view on collaborative human-AI interaction that adoption-focused measures alone cannot provide.