This study examines the effects of generative AI assistance during the prewriting phase of a second-language (L2) summary writing task in a university English course. Fifty-five Japanese EFL students participated in a counterbalanced within-class design involving two summary tasks completed one week apart. In each task, learners used computers to study for 12 minutes a listening script from a homework listening completed the previous week, then wrote a handwritten summary of approximately 60 words from memory. In the AI condition, learners used generative AI tools (e.g., ChatGPT or Copilot) during prewriting; in the non-AI condition, they relied mainly on machine translation. Learners completed two listening-script summaries and switched conditions between tasks.

Summary quality was assessed for content coverage, syntactic complexity, and formal accuracy. Vocabulary learning was measured using a simplified Vocabulary Knowledge Scale administered before and after each task and on a delayed post-test. Post-task surveys and self-reported AI chat logs documented tool use. Learners used AI to identify key ideas, simplify language, generate model summaries, and check drafts, while students relying mainly on machine translation (often Google Translate) showed greater gains in key vocabulary. Full results on writing quality and relationships between AI usage strategies and outcomes will be presented.