Prompt engineering for non-native English learners: A generative AI approach to personalised language feedback

This study explores the effectiveness of ChatGPT-generated feedback on English as a Foreign Language (EFL) students’ writing by comparing three prompt types: general, grammar-focused, and EFL-tailored. It also examines the quality of AI-generated feedback through expert evaluations and explores student perceptions related to clarity, usefulness, and motivation. This study employed an empirical mixed-methods design, combining quantitative expert evaluations and student perception surveys. Thirty-two university students completed a writing task and received feedback generated using each prompt type. Three expert raters assessed the feedback based on relevance, clarity, usefulness, and tone. Inter-rater reliability was confirmed using the Intraclass Correlation Coefficient (ICC). Additionally, a student perception survey was administered in three parts: multiple-choice comparisons, Likert-scale attitudes, and multiple-answer reflections. Results indicate that EFL-tailored feedback was consistently rated highest by both experts and students. Students found this feedback easier to understand, more motivating, and most helpful in improving their writing. The findings underscore the importance of prompt design in generating personalised and effective AI feedback. The study supports integrating ChatGPT into EFL instruction and offers implications for pedagogy, technology use, and future research.