The Impact of Large Language Model Integration on Cognitive Load and Metacognitive Strategies

Abstract

The integration of Large Language Models (LLMs) into educational and professional cognitive processes is transforming the ways in which individuals process information, make decisions, and regulate cognitive activity. Despite the rapid proliferation of LLM-based tools, their impact on cognitive load and metacognitive strategies remains insufficiently explored from an empirical perspective. The present quasi-experimental study aims to examine how the integration of LLMs influences intrinsic, extraneous, and germane cognitive load, as well as the planning, monitoring, and evaluation of one's own cognitive activity. The study involved 124 participants, divided into an experimental group (LLM integration) and a control group. The research instruments included the NASA-TLX cognitive workload scale, the Paas Cognitive Load Scale, and the Metacognitive Awareness Inventory (MAI). Statistical analysis incorporated ANCOVA, Structural Equation Modeling (SEM), and mediation analysis. The results demonstrate a statistically significant reduction in extraneous cognitive load when using LLMs (p < 0.01), accompanied by a redistribution of cognitive resources toward germane cognitive load. At the same time, structural changes in metacognitive monitoring were observed, including the strengthening of planning strategies and external regulation mechanisms. A mediation effect of cognitive load was identified in relation to the influence of LLM use on task performance. The present study contributes to the development of distributed cognition theory and Human-AI Interaction research by proposing an empirically grounded model of cognitive integration of LLMs in professional activity. Thus, the results of structural modeling confirm the hypothesis that the impact of LLMs on performance is primarily mediated through the redistribution of cognitive load and the activation of metacognitive regulatory mechanisms.