<p>Generative artificial intelligence (AI) is rapidly being integrated into writing instruction and assessment, yet evidence on automated scoring, feedback, instructional support, and ethical governance remains fragmented, with limited systematic integration across these dimensions. Following the PRISMA guidelines, this systematic review searched, screened, and analyzed literature published between 2020 and 2024, ultimately including 102 English-language peer-reviewed journal articles across four interrelated areas: automated essay scoring (AES), AI-generated feedback, writing instructional support, and ethical governance. The results indicated that AES generally demonstrated moderate to excellent reliability and moderate to good accuracy, but its performance was influenced by prompt design, parameter settings, task type, and writing quality. AI-generated feedback was generally usable but imperfect. It had advantages in terms of feedback coverage, alignment with evaluation standards, and surface-level corrections, but remained less effective than teacher and peer feedback in identifying higher-order issues, understanding context, and prioritizing key issues. AI supported students’ writing activities during the prewriting, drafting, and revision stages, enhanced motivation and self-efficacy, reduced the burden of surface-level proofreading, and assisted teachers in instructional design, implementation, and evaluation. From an ethical perspective, existing research generally held that AI did not qualify for authorship, its use should be disclosed, and fact-checking, value judgments, and ultimate responsibility should remain with humans. Beyond synthesizing these four strands of evidence, this study further proposes an integrative framework tailored to writing education contexts and provides a theoretical basis for the standardized, stable, and replicable implementation of AI in writing education.</p>