AI-generated level-aligned pre-writing input and EFL reading-to-write development in a Jordanian university context: a longitudinal mixed-methods study

Reading-to-write tasks require EFL learners to comprehend and integrate academic source material into original texts. This longitudinal mixed-methods study examined whether AI-generated, level-aligned input during preparation supported performance among 134 undergraduates in two intact academic writing sections at a Jordanian university. Sections received either AI-mediated input or instructor-scaffolded preparation; AI was prohibited during assessed writing. Performance at three time points was analyzed using repeated-measures ANOVA, alongside reflections, think-aloud protocols, and interviews. Both groups improved, but the AI group gained more, especially in content integration (Group × Time: partial η² = .177; posttest d = 1.58). Learners described benefits for comprehension, synthesis, and planning. However, section-level allocation and potentially unequal synthesis support limit causal inference. The findings position AI as a constrained pre-writing input mediator rather than a drafting tool. Universities should consider stage-specific AI policies that distinguish preparation from independent authorship. Permitting AI for source comprehension, vocabulary clarification, and planning, while restricting its use during assessed writing, may support EFL learners without replacing their work. Guidelines should specify permitted assistance, disclosure expectations, and assessment boundaries. Because the study involved intact classes and potentially unequal support, institutions should pilot such policies cautiously and evaluate their effects on learning and academic integrity.