Metacognition in the Era of Generative AI: Understanding Human Cognitive Engagement and AI Reliance

The growing integration of Generative Artificial Intelligence (GenAI) into human cognitive activities, including problem-solving, decision-making, learning and information processing, has raised important questions about cognitive work allocation, metacognitive control and independent performance. Although GenAI can enhance efficiency by reducing cognitive demands, excessive reliance may weaken users’ engagement in reasoning, verification and independent task execution. This review examines the relationship between GenAI use and human cognition, with particular emphasis on cognitive offloading, appropriate reliance, over-reliance, automation bias and independent performance. Drawing on existing research, it explores how the effects of cognitive offloading depend on task characteristics and the distribution of cognitive responsibility between users and AI systems. The review further considers interventions such as think-first strategies, cognitive forcing functions, reflection, verification prompts and adaptive scaffolding that aim to preserve human cognitive engagement while benefiting from AI assistance. In addition, it examines behavioural telemetry including prompt intent, interaction timing, response editing and verification activities as a potential means of identifying patterns of engagement and reliance during multi-turn human AI interactions. These observable behaviours are treated as indirect indicators rather than definitive measures of internal cognitive processes. The review highlights the importance of evaluating not only task performance with AI assistance but also users’ ability to perform independently when such assistance is unavailable. It identifies a need for integrated frameworks that connect cognitive work allocation, metacognition, observable interaction patterns, reliance calibration and independent performance over time. Such frameworks may support the development of human-centred GenAI systems that improve task effectiveness while maintaining users’ critical thinking, self-regulation and cognitive autonomy.