GenAI-supported task-based EFL learning: equity implications across proficiency levels

Task-based learning (TBL) in English as a Foreign Language (EFL) contexts faces inherent equity challenges, as open-ended tasks yield outcomes shaped by baseline proficiency. Generative AI (GenAI) has been proposed as a supplemental scaffolding layer to alleviate these challenges. However, effective GenAI use draws on proficiency-dependent skills such as prompt formulation and feedback interpretation, raising whether the tool alleviates or amplifies this gap. This 16-week quasi-experimental study examined 129 Chinese university EFL learners stratified into high- and low-proficiency groups and randomly assigned to GenAI-supported or conventional-resource conditions. Equity was assessed through multi-stream convergent analysis spanning learning outcomes, AI interaction patterns, engagement, and motivation. High-proficiency learners achieved significant gains with GenAI support, whereas low-proficiency learners showed no substantial improvement. Six of seven engagement and motivational indicators differed significantly by proficiency, with only intrinsic motivation unaffected, suggesting that GenAI shapes utility-oriented engagement rather than fostering intrinsic interest. Although the overall Group × Proficiency interaction did not reach significance, convergent evidence across achievement trajectories, AI interaction patterns, and motivational experiences pointed toward unequal benefit distribution. These findings challenge the assumption that open-access GenAI tools inherently democratize second-language learning and call for capability-oriented instructional redesign to mitigate proficiency-based gaps.