<![CDATA[Mastery of vocabulary occupies a central position in global literacy development; however, the restricted time allocation for English instruction in primary schools frequently limits student learning outcomes, especially among early-grade learners. This study investigates whether a statistically meaningful difference exists in English vocabulary achievement between students learning through Creta Class, an AI-powered adaptive platform, and those in conventional instruction settings. The study adopted a quasi-experimental approach using a non-equivalent control group design, with 48 second-grade students as participants, distributed into an experimental group (n = 25) and a control group (n = 23). A 20-item multiple-choice instrument covering five vocabulary domains served as the data collection tool. Statistical analysis using the independent samples t-test revealed no significant difference between groups (p = 0.105 > 0.05). Nevertheless, descriptive findings indicated a moderate learning gain in the experimental group (N-gain = 0.37), compared to a low gain in the control group (N-gain = 0.19), with a moderate effect size (Cohen's d = 0.478). These results suggest that AI-powered adaptive learning carries meaningful practical value in supporting individualized learning experiences and real-time feedback delivery, even in the absence of statistical significance. The study underscores the necessity of sound instructional design and extended intervention duration to optimize AI-integrated learning in elementary education contexts.]]>
