Generative artificial intelligence (AI) is being increasingly integrated into educational practice, changing the dynamics of writing instruction and feedback. This study examines the relative effectiveness of writing instruction based on (1) instructor feedback versus (2) generative AI-assisted feedback on undergraduate ESL learners' academic writing skill, revision quality and self-efficacy. We utilized an explanatory sequential mixed-methods design with 100 undergraduate students randomly assigned to AU (control, traditional instructor feedback) or experimental (AI-assisted spelling/grammar feedback) groups over an eight-week intervention. Analytic rubric was used to evaluate writing performance, whereas the questionnaire was administered to calculate writing self-efficacy. Qualitative data were obtained using semi-structured interviews. Illustrative results show some additional improvements with AI‐assisted feedback on both quality of revision and writing confidence (compared to instructors) but instructor feedback was still better at supporting higher‐order argument development. The results emphasize the educational potential that generative AI can provide as an additional feedback resource in a CALL context.