Exploring Postgraduate Students' Generative AI Use in English Academic Writing: Practices, Challenges and Pedagogical Implications

Aims: This study investigated the use of generative artificial intelligence (GAI) by postgraduate students in English as a foreign language (EFL) thesis introduction writing, focusing on their interaction practices, rhetorical and functional writing difficulties, and relationships between these difficulties and observable GAI use. Study Design: An exploratory, classroom-embedded study combining qualitative analysis with descriptive statistics. Place and Duration of Study: A postgraduate thesis writing course at a Chinese university, September–December 2025, with data drawn primarily from the first five weeks. Methodology: The study analysed 35 students' thesis introduction drafts and corresponding GAI interaction records, supplemented by stimulated-recall interviews with three participants. GAI-use practices were examined in terms of interaction patterns, task focus, and text-generation responsibility. The drafts were analysed with reference to the Create a Research Space model and functional adequacy, followed by a cross-source analysis. Results: Students used GAI for diverse writing tasks and differed in their allocation of text-generation responsibility, but sustained interaction around individual writing problems was limited. Although most drafts contained recognisable rhetorical moves, difficulties persisted in literature synthesis, research-gap construction, and argumentative coherence. Cross-source analysis further revealed difficulties in identifying genre-specific writing problems, formulating targeted GAI requests, and evaluating generated content against relevant evidence and rhetorical expectations. Conclusion: Meaningful GAI-assisted academic writing requires more than frequent interaction or control over text generation. Genre-specific diagnostic criteria and process-oriented support are needed to promote purposeful GAI use and evidence-based authorial judgement.