Using AI to adapt reading input for mixed-proficiency EFL learners

Mixed-proficiency EFL classrooms complicate the provision of level-appropriate reading input. This study examined whether AI-based text adaptation improves reading comprehension for learners at different proficiency levels. Grounded in Krashen's Input Hypothesis and Cognitive Load Theory, the study used a 2 × 2 mixed factorial design (text version: original vs. AI-adapted; proficiency: Intermediate-Low vs. Advanced-Low) to test performance on inference-making and reading-for-specific-information tasks. Forty-eight university EFL students, classified using ACTFL guidelines, completed comprehension tasks using original and AI-adapted versions of the same source texts, with text-version order counterbalanced using two parallel test forms to reduce order effects. Intermediate-Low learners scored significantly higher on adapted than on original texts ( p < .001), whereas Advanced-Low learners showed comparable performance across versions. A significant text-by-proficiency interaction ( p < .001) suggests that AI-assisted adaptation can support differentiated reading instruction by improving access to comprehensible input for lower-proficiency learners without disadvantaging higher-proficiency learners, while providing a transparent workflow for proficiency-aligned adaptation.