Generative AI Dependence and Critical-Thinking Performance Among University Students: The Mediating Role of Cognitive Offloading - A Structured Evidence Synthesis

Abstract

Background: Generative artificial intelligence (GenAI) can reduce the cognitive effort required for information retrieval, summarization, drafting and problem solving. Cognitive offloading is not inherently harmful, but dependence may emerge when learners delegate the very processes that higher education is intended to develop, including interpretation, evaluation, inference and justification.
Objective: To synthesize evidence available on the association between GenAI dependence and critical thinking among university students and to evaluate cognitive offloading as a plausible mediating mechanism.
Methods: A structured evidence synthesis was conducted using peer-reviewed Q1 literature in educational technology, learning sciences, psychology and cognitive science. Direct studies of AI dependence and critical thinking were prioritized, together with experimental studies of ChatGPT-supported learning and foundational cognitive-offloading research. Because the available studies used heterogeneous constructs and only a small number directly estimated mediation, no pooled effect size was calculated.
Results: The strongest direct evidence came from a 2025 study of 580 university students in which AI dependence correlated negatively with critical thinking (r=-0.35) and positively with cognitive fatigue (r=0.51). Cognitive fatigue partially mediated the relationship: AI dependence predicted fatigue (beta=0.51), fatigue predicted lower critical thinking (beta=-0.33), and the direct AI-dependence effect decreased from beta=-0.35 to beta=-0.19 after inclusion of the mediator. Experimental evidence produced a complementary finding. In a randomized study of 117 university students, ChatGPT support improved essay revision but did not significantly improve knowledge gain or transfer and altered self-regulated learning processes in a pattern interpreted as possible metacognitive laziness. Conversely, a study of 213 Italian students found that active engagement with ChatGPT was positively associated with complex critical-thinking performance, and a 2025 meta-analysis of 69 experimental studies concluded that ChatGPT interventions improved higher-order thinking while reducing mental effort. A time-lag study of 465 preservice teachers further showed that cognitive offloading can positively mediate academic outcomes when routine tasks are externalized while higher-order work remains with the learner.
Conclusion: GenAI dependence is associated with weaker critical thinking, but the mechanism is not simple tool use. Cognitive offloading appears beneficial when it removes routine cognitive burden and harmful when it substitutes for analysis, monitoring and independent reasoning. Universities should therefore design assessment and AI literacy practices that preserve productive cognitive effort while allowing strategic offloading.