Evaluating Generative AI in Postgraduate Assessment Through Writing Quality, Accuracy and Pedagogical Implications

Generative AI tools such as ChatGPT and Microsoft Copilot are increasingly used in student academic writing, raising questions about learning, integrity, and assessment standards. This paper presents an autoethnographic case study in which Microsoft Copilot was used to generate a 4,500-word draft for a postgraduate assignment on “pig-butchering” cryptocurrency scams and the effectiveness of UK law in addressing them. The draft was evaluated for writing quality, factual accuracy, legal analysis, and evidential support. The findings show that Copilot produced fluent, well-structured text at an appropriate academic level, but key limitations were identified: approximately one-third of references were fabricated or untraceable, several citations contained errors, and the analysis was often descriptive rather than critically engaged. Legal discussion identified relevant UK statutes but offered limited, weakly evidenced case examples. The study also reflects on how iterative prompting shaped outputs and the level of human verification required. Implications for pedagogy and assessment are discussed, with recommendations for embedding AI literacy and transparent, responsible AI use in higher-education curricula.