Deriving Metacognitive Scaffolding Design Principles and an Integrated Model for Generative AI–Supported Major Exploration

As flexible major selection systems, including undeclared admission, become increasingly widespread in higher education, learners' capacity for self-directed academic and career planning has become more important than ever.Although the widespread adoption of Generative AI (GenAI) has improved the efficiency of information acquisition, it also raises concerns about intensified cognitive offloading, whereby learners become overly dependent on AI-generated responses without engaging in critical thinking.To address this issue, this study proposes an AI-based integrated metacognitive scaffolding model that enables learners to use AI not merely as an information provider, but as a tool for expanding their thinking.To this end, an integrative literature review explored the theoretical relationship between metacognition and scaffolding, and a cross-mapping analysis was used to derive stage-specific design principles.In addition, a scenario-based walkthrough examined the logical applicability of the proposed model.The analysis suggests that AI-supported major exploration can be conceptually organized around the role of an Inquiring Coach.The proposed model organizes the learning process into three metacognitive phases-Planning, Monitoring, and Evaluation-and specifies stage-specific strategies, including delaying immediate answers, detecting inconsistencies and generating counter-questions, and promoting reflective self-explanation.The model also incorporates a Fading mechanism as a design principle intended to support the gradual development of learners' selfdirected exploration.This study provides conceptual design guidelines for AI-supported career education systems.It proposes a theoretical basis for shifting attention from information-delivery efficiency to designing interactions that support learners' thinking.However, the model's effects on learners' metacognition or career decision-making were not empirically tested.