Generative AI in K-12 creative writing instruction: A systematic review and narrative synthesis

Generative AI is rapidly entering writing instruction in public education; however, a systematic, quality-appraised synthesis of its effects on K-12 students' creative writing is lacking. This review synthesises and critically appraises the empirical evidence, using a scaffolding-versus-substitution framework to distinguish AI-assisted performance from the development of students' own writing. We searched seven databases without language restrictions for 2020-2026, following PRISMA 2020. Risk of bias and methodological quality were assessed with ROBINS-I, RoB 2 and CASP, the certainty of evidence with GRADE and GRADE-CERQual, and findings were synthesised without meta-analysis (SWiM), stratified by design. Seventeen studies met the criteria (total N = 736). All six comparisons of AI-supported with non-AI instruction favoured AI on text quality or creativity, but in every study the scored texts were produced with AI access or under unreported conditions; no study assessed AI-free or delayed writing performance. Effects on writing-related affect were mixed, and no intervention reported a planned withdrawal of AI support. Observational process data indicated that students used AI input to develop their ideas when they retained control, but that AI could also steer the narrative or supply text adopted with little reworking. No study was at low risk of bias, and the certainty of evidence was very low for all quantitative outcomes. The evidence therefore describes what students produce with AI rather than what they learn to do without it; studies with AI-free post-tests, delayed transfer measures and planned fading of support are needed. The protocol was pre-registered (PROSPERO: CRD420261303257).