In this talk we reflect on two years of incorporating generative AI tools, such as ChatGPT, into the assessment practices of second year undergraduate specialist mathematics modules. These include a problem solving and communication module, assessed through critiques and multimedia productions, and a
mathematical statistics module assessed through open-ended statistical analysis with explicit comparison to LLM output. These have provided fertile ground for evaluating the affordances and limitations of AI in supporting mathematical learning and expression.
Drawing on student voice, submissions and structured feedback, we examine how critical comparison with LLM output and AI-assisted drafting, scripting, and idea generation has influenced students’ engagement, creativity, and technical clarity. While generative AI offers support in refining language and breaking down
complex concepts, challenges emerge in over-reliance, missed opportunities for genuine mathematical reasoning, and the masking of individual voice.
We also consider institutional requirements for transparency and ethical use of AI, and how these shape students’ attitudes and reflective practices. Grounded in the principles of mathematical communication, we argue that although AI tools can scaffold the technical expression of ideas, the core pedagogical value
remains in nurturing personal insight, mathematical authenticity, and adaptive problem-solving skills.
We conclude with practical recommendations for integrating AI in a way that supports, rather than replaces, student ownership of mathematical narratives.
