The outcome-based education (OBE) framework emphasizes rigorous constructive alignment among learning outcomes, instructional activities, and assessment criteria. Within English for Specific Purposes, particularly Commentary Writing, providing detailed, genre-specific and pedagogically coherent teacher feedback remains challenging. The emergence of artificial intelligence large language models (AI LLMs) offers a potential solution by enhancing feedback quality and enabling an efficient scaffolding approach. This study proposes an AI-empowered teacher scaffolding model in commentary writing teaching, where AI specifically focuses on genre features, informational completeness, and linguistic quality, while the teacher acts as a pedagogical mediator, strategically designing prompts, critically evaluating AI output, and transforming it into personalized and outcome-oriented feedback. It also explores the design principles of this human-machine collaborative feedback system in aligning with OBE outcomes, and students’ perceptions of its efficacy. This quasi-experimental intervention study compares an experimental group (AI-mediated scaffolding) with a control group (traditional teacher feedback). Students’ pre-revision text (version 1) and post-revision text (version 2) were collected and analyzed. Quantitative analyses included multi-dimensional textual comparisons. Qualitative data from teacher logs and student interviews helped to assess the instructional process and subjective experiences. Findings showed this AI-empowered scaffolding model significantly enhanced students’ ability to master news genre conventions and overall writing competencies compared with traditional feedback. Teachers’ mediation of AI feedback ensured alignment with OBE learning outcomes and fostered learners’ autonomy and critical evaluation skills regarding AI suggestions. This study presents an effective methodology for incorporating LLMs into journalism commentary writing instruction while maintaining pedagogical integrity.