<![CDATA[Limited lexical depth and the difficulty to retain words after teaching are issues for junior high school students who learn to speak English as a foreign language. This study investigated vocabulary learning and student perceptions of a Gemini AI-supported digital poster project in a Grade 8 classroom in Palembang, Indonesia. A mixed-methods quasi-experiment was conducted on two intact classes (N = 48; 24 per group) in six meetings. The experimental group explored meanings, context and associated meanings using Gemini AI and produced Canva posters through repeated practice, peer interaction and teacher guidance. The 30 item vocabulary assessment was done before instruction and immediately afterward and two weeks later. Fifteen item questionnaires and six semi-structured interviews were used to extract perception information. The experimental and control mean post-test was 84.37 and 71.04 respectively. The two-week delayed means were 80.92 and 66.13 as compared to the final results from the summaries. The post-test mean difference was 13.33 (95% CI [9.34, 17.32], p < .001) and the delayed difference was 14.79 (95% CI [10.65, 18.93], p < .001). These cross-sectional tests do not establish a group-by-time effect. Students generally described interest, contextual support and creativity as well as connectivity problems, uneven digital skills and a dependence on AI. The instructional package was associated with higher immediate and two-week delayed scores, but its individual components cannot be separated. We need to be mindful of teacher monitoring, checking AI responses, privacy protection and the need for AI literacy and education.]]>
