Leveraging GPT-Based Large Language Models for Personalized Feedback in ESL Writing: Advancing Automated Writing Evaluation and Learner Development

This study explores the application of transformer-based Large Language Models (LLMs), specifically GPT models, for providing personalized feedback in English as a Second Language (ESL) writing. Traditional automated writing evaluation (AWE) systems often fail to deliver nuanced, learner-specific guidance, limiting their pedagogical value. Using a dataset of 1,500 annotated intermediate-level ESL essays, the research compares GPT-generated feedback with rule-based and statistical AWE systems across error detection accuracy, feedback clarity, personalization, and learner engagement. Quantitative results demonstrate that GPT-based models outperform conventional systems in detecting grammatical, lexical, and stylistic errors while offering actionable revision suggestions. Qualitative analysis indicates that GPT-generated feedback enhances learner comprehension, encourages reflective revision, and supports self-directed learning. The study also discusses challenges such as overcorrection, first-language interference, and pedagogical alignment. Findings suggest that GPT-based feedback systems have the potential to transform automated writing evaluation and improve ESL learning experiences, offering scalable, adaptive, and pedagogically informed support for learners and educators alike. Keywords: Large Language Models, GPT, ESL writing, personalized feedback,