Tracing student engagement in dialogues with GenAI in assessment: A semester-long exploratory study

There are many unsupported claims in the media about how students use generative AI (GenAI) to support their assessment tasks, often labelling such use as inherently unethical. In response, this paper presents a semester-long investigation into 11 university students’ engagement in dialogues with GenAI for their graded assignments, using authentic chat records as the primary data. The study identified four student engagement patterns with GenAI – behavioural, (meta)cognitive, emotional and agentic. While less common, agentic engagement appeared in 12% of the dialogues where students intentionally challenged GenAI’s output and integrated their own understanding of the assessment to direct its responses. Such engagement shows that students’ GenAI use does not necessarily mean they avoid critical thinking or deep learning. We argue that promoting agentic engagement requires more than teaching students how to write better prompts but more deliberate effort to help students reflect on the purpose and process of their GenAI use.