English as a Foreign Language (EFL) writing classrooms are increasingly adopting generative Artificial Intelligence (AI) tools, which provide additional scaffolding but also raise important questions about their impact on learners' critical thinking. Applying a quasi-experimental mixed-methods design, this study compared the writing performance and critical thinking of 94 undergraduate EFL learners who were assigned in equal numbers to an AI-supported instruction group (n = 47) and a teacher-mediated instruction group (n = 47). The 12-week intervention was guided by Paul and Elder’s Critical Thinking Framework for teaching, learning, and assessment. Data collection instruments included pre- and post-intervention writing tests, a critical thinking rubric, reflective journals, AI interaction logs, and semi-structured interviews. Quantitative data were analyzed using independent-samples t-tests and one-way ANOVAs, with Cohen’s d used to estimate effect sizes; qualitative data were analyzed thematically. The results showed that the AI-supported group scored significantly higher than the teacher-mediated group on the post-intervention writing task, t(92) = 2.97, p = .004, and completed the task significantly faster, t(92) = 8.23, p < .001. Although both groups improved their writing performance over the intervention period, the within-group gain was larger in the AI-supported group (d = 2.12) than in the teacher-mediated group (d = 1.32). However, post-test critical thinking scores did not differ significantly between groups, t(92) = 0.69, p = .49. Thematic analysis further showed that, over the course of the semester, learners shifted from passively accepting AI suggestions to critically evaluating AI-generated content. Overall, the results indicate that AI-mediated instruction improved writing quality and efficiency without any evidence of a detrimental effect on learners’ critical thinking; instead, learners increasingly evaluated AI-generated content critically before incorporating it into their writing.