Narrative review: GenAI cognitive offloading can boost immediate performance but shows a performance-learning dissociation, and which functions are offloaded matters more than whether AI is used

The review's key finding is that whether GenAI is used at all is less critical than which functions are offloaded, which are maintained, and the extent to which users remain epistemically responsible for the resulting output. Prior discussion often stayed within a binary 'AI helpful or harmful' frame; this work reframes the question as one of function allocation and epistemic responsibility. Based on a structured narrative search for experimental, meta-analytic, and mixed-methods studies on memory, learning, critical thinking, creativity, perceived effort, trust, and dependence on GenAI, complemented by cross-sectional and qualitative studies.

GenAI-supported cognitive offloading can reduce the cognitive capacity required and support high performance across a variety of tasks, including memory-based and GenAI-supported academic tasks. This aligns with foundational cognitive-offloading research showing that external support can boost performance when support is available. The review synthesizes foundational cognitive-offloading literature together with GenAI-related experimental and meta-analytic evidence.

The same cognitive offloading can in some cases reduce the quality of subsequent encoding, and recent GenAI studies show a consistent performance-learning dissociation: gains from support available at the time typically carry over to later unsupported performance only slightly, not at all, or even negatively. This dissociation separates 'performing well now' from 'actually learning,' and is the review's key contribution to the GenAI cognition discussion. The review cites experimental and meta-analytic work on memory, learning, and transfer, and notes the pattern recurs across recent GenAI studies.

Studies of support provided under instructional control indicate that retaining learner-generated nodes or restricting direct-answer substitutes for generators can support later independent performance better than unlimited support, while the independent contribution of explanation, verification, revision, and fading support remains to be determined. This offers actionable direction for education while explicitly marking unresolved research questions. Grounded in studies of support provided under instructional control, though the review states the independent contribution of each element is not yet determined.

This review is aimed at educators, knowledge workers, tool designers, and researchers, and applies to contexts of GenAI-assisted memory, learning, critical thinking, and creativity tasks. Its conclusions are meant to guide how support is controlled in instructional design, how tools are designed to retain learner-generated content, and how longitudinal research is conducted, rather than to provide a general diagnosis of individual cognitive ability.

Readers should still watch: the independent contribution of explanation, verification, revision, and fading support remains to be determined; emerging dependence research correlates with academic stress, performance expectations, anxiety, and self-efficacy, but much of that evidence is cross-sectional or qualitative; current evidence does not support claims of irreversible cognitive decline, yet long-term causal effects still require longitudinal testing. In addition, the loaded text is an incomplete version lacking the body, figures, and specific statistics, so the specific effect sizes and sample characteristics of the included studies cannot be summarized.