This meta-analysis synthesizes evidence on the effects of ChatGPT-assisted instruction on English writing in Korean EFL contexts. A systematic search in RISS yielded 401 records; after screening under PRISMA procedures, 12 studies reporting writing outcomes were included and analyzed. Effect sizes were computed as Hedges’ g and pooled using random-effects models. The random-effects mean indicated a medium-tolarge improvement in writing (g = 0.6867). Mixed-effects subgroup analyses detected no statistically significant differences by experimental design (pre-experimental vs. quasiexperimental), educational level (tertiary vs. secondary), or region (metropolitan vs. others). Meta-regression trends for duration (weeks, positive) and weekly intensity (sessions/week, negative) were suggestive but not statistically significant. Overall, ChatGPT-supported writing instruction yields meaningful gains and shows consistent effects at both secondary and tertiary levels; in practice, it is well suited to complement teacher feedback. Future studies should expand samples in secondary and nonmetropolitan settings and employ stronger designs to clarify how duration and intensity shape outcomes.