This study examined how writing proficiency and grouping methods influence the effectiveness of trained peer feedback. Employing a 2×2 between-groups design, 80 EFL learners from a Chinese university were systematically categorized into four groups (20 per group) based on writing proficiency (high/low) and grouping method (homogeneous/heterogeneous). During the initial semester, students received training in implementing peer feedback within their writing classes. In the second semester, each experimental group received peer feedback across multiple assignments. Standardized assessments of writing proficiency, writing self-efficacy, and writing self-regulated learning were conducted at the beginning and end of the second semester, while classroom experience was evaluated only post-intervention. For students with high writing proficiency, peer feedback proved most effective in homogeneous groups, resulting in significantly enhanced writing development, self-efficacy, and autonomous motivation. Conversely, students with low writing proficiency benefited most from peer feedback in heterogeneous groups, leading to notable improvements in writing development and self-efficacy. The study advises against using groups of two and suggests using AI assistance for lower-ability students in peer reviews, although further clarification on these recommendations may be warranted.