Abstract
Generative AI is increasingly used in learning environments, yet sustained interaction does not necessarily indicate cognitively engaged learning. This exploratory study examined how conversation openings related to cognitive engagement levels and the support orientations reflected in student messages during student-AI interactions. Cognitive engagement was classified by an adapted ICAP framework, openings were organised into a descriptive typology, and semantic thematic analysis was conducted on student messages. Openings oriented towards strategy guidance, conceptual explanation, error understanding, and resource review feedback were more often associated with moderate or deep engagement, whereas resource-evaluation openings were often associated with longer interactions despite limited cognitive depth. Semantic patterns revealed overlapping orientations towards questioning, improvement, examples, feedback, and concepts, spanning clarification, answer checking, revision, application, and task-focused support. These findings distinguish interaction persistence from cognitive engagement and suggest that conversation design should support students to articulate their reasoning and co-construct knowledge with AI, rather than prioritising continued interaction alone.
Keywords
Cognitive engagement, ICAP framework, Student-AI interaction, Semantic analysis, Conversation openings, Educational chatbot, Exploratory study.
