This quasi-experimental study investigated the effect of guided ChatGPT integration on the writing performance of Iraqi EFL university students. A sample of 120 undergraduate students from two faculties at a public Iraqi university was assigned to an experimental group (n = 60) that used ChatGPT-assisted writing tasks and a control group (n = 60) that followed conventional teacher-centered writing instruction. Both groups took an identical pretest and posttest writing task scored with a standardized analytic rubric; an attitudes questionnaire and interviews complemented quantitative data. Results show that the experimental group improved significantly more than the control group on overall writing scores (experimental posttest mean = 68.4, SD = 8.6; control posttest mean = 57.1, SD = 10.7; independent-samples t(118) = 6.38, p < .001, Cohen’s d ≈ 1.16). Paired comparisons within the experimental group also showed large gains from pretest (Mpre = 54.8, SD = 9.9) to posttest (Mpost = 68.4, SD = 8.6; paired t ≈ 11.1, p < .001). Questionnaire and interview data indicated positive attitudes: students reported increased confidence, faster revision cycles, and improved organization and vocabulary. The paper discusses pedagogical implications, limitations, and recommendations for integrating large language models (LLMs) into EFL writing instruction.