AI dependency and academic self-efficacy among university students: a multivariable cross-sectional study

Background Generative artificial intelligence (AI) has become increasingly integrated into university education, raising questions about whether frequent reliance on AI is associated with the academic confidence and performance of students. However, evidence on AI dependency and academic self-efficacy remains limited and inconsistent, particularly among university students in Saudi Arabia. This study examined the association between AI dependency and academic self-efficacy. Methods We conducted a cross-sectional web-based survey among university students recruited from three public universities located in Riyadh, Saudi Arabia, between 20 May 2026 and 15 June 2026. Participants completed measures of AI dependency, academic self-efficacy, daily AI use, and sociodemographic and academic characteristics. AI dependency was assessed using a nine-item measure adapted from the AI Dependence Scale, while academic self-efficacy was assessed using the five-item General Academic Self-Efficacy Scale. Multivariable linear regression was used to examine the association between AI dependency and academic self-efficacy, adjusting for daily AI use, gender, age group, year of study, and college. Results The analysis included 227 students: 59.9% of them were from health sciences programs, 11.5% were from engineering and technology programs, and the remaining participants were distributed across other academic fields. The AI dependency scale demonstrated high internal consistency (Cronbach's α = 0.837; McDonald's ω = 0.940) similar to the academic self-efficacy scale (Cronbach's α = 0.943; McDonald's ω = 0.945). Higher AI dependency was significantly associated with higher academic self-efficacy after adjustment for potential confounders (standardized β = 0.25; 95% CI, 0.11–0.39; P < 0.001). The overall model explained 44% of the variance in academic self-efficacy ( R 2 = 0.44; adjusted R 2 = 0.40). Conclusions Higher AI dependency was associated with greater academic self-efficacy among university students. Given the cross-sectional design, self-reported measures, and potential for common method bias, this association should not be interpreted causally. Furthermore, longitudinal and experimental studies are needed to clarify the temporal relationship between AI dependency and academic self-efficacy. Because the participants of this study were recruited from only three public universities in Riyadh and health sciences students were predominant, the findings should not be generalized directly to all university students in Saudi Arabia.