Kwon Taehyun This study compared the effects of feedback generated by human evaluators and ChatGPT on high-school students’ revision performance in argumentative writing to investigate the educational effectiveness and characteristics of generative AI-based writing feedback. A quasi-experimental pretest-posttest design was used with 154 second-year high school students to examine quantitative effects and identify qualitative features through a learner perception survey. The results showed that the effect size of AI feedback was slightly larger than that of human feedback, and the two-way ANOVA revealed a significant main effect only for writing proficiency level. According to qualitative analysis, AI feedback demonstrated strengths in specificity, feasibility, systematicity, consistency, and a focus on improving linguistic expression, but it also had limitations related to stereotypical patterns and insufficient support for metacognitive engagement. These results suggest that generative AI feedback can be used effectively as a tool for initial revision, whereas a hybrid approach that combines it with teacher feedback is required to address higher-order argumentative elements and promote metacognitive processes.