This study examines two AI-driven written feedback tools—ChatGPT and AI Digital Textbooks (AIDT)—by comparing their effects on Korean middle school students’ English writing performance and perceptions. Over six weeks, fifty seventh graders completed six expository writing tasks and received feedback from either ChatGPT or AIDT. Pre- and post-tests and questionnaires assessed changes in writing performance and affective factors, including motivation, self-efficacy, metacognition, and anxiety. Results showed significant improvement in both groups with large effect sizes, although overall post-test scores did not differ. Subcomponent analyses revealed that ChatGPT better supported content and idea development, while AIDT was stronger in grammar and cohesion. ChatGPT users also reported higher motivation, self-efficacy, and perceived usefulness. Overall, both tools supported L2 writing, but with distinct pedagogical strengths: AIDT promoted accuracy through structured, rule-based feedback, whereas ChatGPT provided adaptive, dialogic feedback encouraging higherorder thinking. These findings suggest that tool selection should align with learners’ proficiency levels and instructional goals across different stages of writing.