AI tool features and their influence on social presence and engagement in foreign language learning

The design of AI-powered educational tools requires careful attention to both technical functionality and social-affective dimensions, yet most prior research has examined these as a single construct. This study investigates how AI tool design features specifically sociability and functional effectiveness shape social presence and student engagement in English Language Teaching (ELT). Using a convergent mixed-methods design, data were collected from 100 undergraduate students enrolled in the English Education program at Universitas Negeri Makassar, Indonesia, who had used two AI tools (Cici AI and Tutor Lily) for a minimum of one semester. A 20-item Likert-scale questionnaire measuring three theoretically grounded dimensions Social Presence, Engagement, and Effectiveness was administered in English and analysed alongside open-ended qualitative responses. Internal consistency was excellent across all subscales (Cronbach’s α = 0.976–0.986). Pearson and Spearman correlations revealed that Social Presence was moderately associated with Engagement ( r = 0.446, ρ = 0.462, p < 0.001, 95% CI [0.273, 0.591]), constituting a medium-to-large effect (Cohen, 1988), while Engagement and Effectiveness were strongly correlated ( r = .670, ρ = 0.659, p < 0.001). Critically, Social Presence and Effectiveness showed a near-zero, non-significant relationship ( r = − 0.028, p = 0.780, 95% CI [− 0.223, 0.169]), confirmed by post-hoc power analysis (G*Power 3.1; detectable r ≥ 0.28 at n = 100, α = 0.05, power = 0.80) as a genuine null finding. A multiple regression model further demonstrated that Social Presence and Effectiveness jointly predicted 66.5% of the variance in Engagement (R² = 0.665, F (2, 97) = 96.33, p < 0.001). These findings indicate that social presence and technical effectiveness operate through complementary yet distinct pathways in AI-mediated language learning, each requiring deliberate and independent design investment. Qualitative themes including enhanced learner confidence, increased motivation through personalized challenges, and challenges related to contextual understanding and unnatural interaction corroborated and extended the quantitative results. The study contributes to Human-Centered AI design frameworks by demonstrating that affective and functional dimensions of AI tools must be developed as parallel, non-redundant design priorities to effectively support EFL learners.