Effects of Three AI-Enhanced Instructional Packages on Academic Vocabulary Outcomes in Chinese EFL Postgraduate Students: A Cluster-Randomized Controlled Trial

This cluster-randomized controlled trial compared the effects of three AI-enhanced instructional packages—a Wordnik-based contextualized package, an Edmodo-based formative-assessment package, and a Duolingo-based gamified package—with traditional instruction on academic vocabulary application, retention, and acquisition among 572 Chinese postgraduate EFL students. Outcome-specific ANCOVAs, adjusted for baseline performance, showed that the Wordnik-based contextualized package produced the largest improvements over the control in vocabulary application (adjusted mean difference = 34.56), retention (11.32), and acquisition (12.76), with all Holm-adjusted p values < .001. The Edmodo-based formative-assessment package also demonstrated significant advantages across all three outcomes, whereas the Duolingo-based gamified package showed a modest improvement in acquisition (adjusted mean difference = 2.08, Holm-adjusted p < .001) but no significant advantage in retention ( p = .195). Qualitative findings supported differences in perceived contextual relevance, feedback quality, engagement, and academic alignment but did not establish causal mechanisms. Because each pedagogical modality was implemented through a single platform, the estimated effects represent complete intervention packages rather than isolated platform or modality effects. These findings provide evidence to inform the design of AI-supported academic vocabulary instruction for postgraduate EFL learners.