<![CDATA[Oral proficiency is the area in which Indonesian English majors most often fall short, and the arrival of conversational artificial intelligence has widened access to speaking practice without settling the question of who actually benefits from it. Intervention studies report average gains yet say little about the learner attributes behind them, so departments now weighing AI literacy modules are deciding without evidence on the learners themselves. This study examined how far AI literacy, speaking self-efficacy, digital engagement, and speaking anxiety predict perceived EFL speaking performance, both jointly and individually. A quantitative, non-experimental, cross-sectional correlational survey was conducted with 178 undergraduates in an English Language Education programme at a university in South Sulawesi, drawn by proportionate stratified random sampling with year of study as the stratifying variable. Five Likert scales totalling 63 items measured the four predictors and the outcome, all of them meeting the conventional item and reliability criteria. The questionnaire was administered online over four weeks in mid-semester, and the data were analysed in IBM SPSS Statistics 27 through descriptive statistics, Pearson correlations, classical assumption testing, and simultaneous multiple regression. The four predictors together accounted for 54.6 per cent of the variance in perceived speaking performance, F(4, 173) = 51.940, p < .001. Speaking self-efficacy made the largest unique contribution (β = .329, sr² = .068), followed by speaking anxiety (β = -.252, sr² = .046) and digital engagement (β = .217, sr² = .031). AI literacy remained significant while contributing least (β = .175, sr² = .018), despite a zero-order correlation of .542 with the outcome. Because the design is correlational, that contrast is read as overlap rather than as a causal pathway: the share of variance uniquely attributable to AI literacy shrinks once confidence, practice, and apprehension are taken into account. Departments planning AI literacy modules are therefore better advised to situate them within a wider affective and behavioural strategy than to adopt them as a stand-alone remedy for weak oral performance.]]>
Copyrights © 2026
