Can't stop, won't stop: continuance ChatGPT usage—enhancing performance or creating dependency?

The adoption of ChatGPT in higher education presents a paradox: it enhances academic productivity while creating cognitive dependency risks that may compromise critical thinking. This dual-outcome phenomenon remains poorly understood, particularly in developing countries like Indonesia where adoption occurs without institutional guidance. This study examines how UTAUT factors influence satisfaction and continuance intention toward ChatGPT, and investigates whether continuance intention simultaneously leads to positive outcomes (academic performance) and negative outcomes (AI dependency). Using a quantitative approach with 714 respondents from Indonesian universities (454 undergraduate, 260 postgraduate), this study employs PLS-SEM to test an extended UTAUT-ECM framework from the I-PACE perspective. Data were collected through stratified convenience sampling from four major educational cities. The results reveal three paradoxical patterns: (1) effort expectancy positively affects satisfaction but negatively affects continuance intention, contradicting traditional UTAUT; (2) continuance intention simultaneously increases academic performance (β = 0.291) and AI dependency (β = 0.361), with dependency effects being stronger; and (3) social influence is non-significant across all groups, indicating bottom-up adoption amid weak institutional support. Educational level differences emerge, with undergraduates showing stronger habit formation while postgraduates display selective satisfaction patterns. This study demonstrates that traditional UTAUT assumptions do not fully apply in generative AI contexts: ease of use becomes counterproductive, and social influence is irrelevant without institutional support. The findings emphasize differentiated interventions—habit management for undergraduates and skill diversification for postgraduates—alongside dual institutional strategies optimizing educational benefits while implementing dependency safeguards.