Comparative Analysis of Application Modes and Feedback Effects of Large Language Models in L2 Writing

Traditionally, second language(L2)writing feedback has long depended on teachers' manual error correction. In practice, however, typical classroom settings often prevent such teacher-led feedback from being implemented effectively. This is why large language model (LLM)-based artificial intelligence (AI) feedback has rapidly emerged as a widely adopted alternative in recent years. Even so, existing relevant research does not produce consistent and reliable findings. At the same time, newly released AI tools on the market have made many teachers unsure about these tools' selection and practical classroom integration. This literature review divides LLM-generated feedback for L2 writing into four core types: holistic assessment, sentence-by-sentence error correction, structured revision suggestions, and multi-turn interactive dialogue.It analyzes six points: feedback type, learning effectiveness, learner autonomy, learner proficiency, and the strength of empirical research support. The results show that there is no specific feedback mode that can be used in every scenario, because each mode needs to be used with different learner groups. The review also finds some key limitations in existing review work. It provides actionable ways for both teachers and AI tool developers, gives suggestions to build unified criteria for LLM feedback, and notes that in second language writing instruction, there is a need to research how to combine human teacher feedback with LLMs' feedback flexibly.