Reflexive AI Literacy: A _Phrasome_-Aware Framework for Generative AI Use and Judgment Formation

Generative AI is increasingly used not only to produce text, but also to assist users in forming judgments, interpreting information, drafting formal documents, and deciding how to act. This paper argues that the central risk in such use is not merely misinformation, but AI-generated confidence: the tendency for fluent, coherent, and context-sensitive language to be mistaken for understanding, authority, or reliable judgment. Building on the concept of the Phrasome, understood as a linguistic structure through which non-subjective language can be misrecognized as the presence of a subject-like interlocutor, this paper proposes a reflexive framework for AI literacy. The framework identifies three foundational forms of user awareness: generative-system awareness, non-subject and non-authority awareness, and awareness of systemic tendency and initiative limitation. It then maps key risks in AI-mediated conclusion formation, including linguistic misrecognition, epistemic unreliability, task and semantic misclassification, system-specific tendency, and conclusion over-stabilization. In response, the paper proposes seven reflexive procedures for generative AI use: question or claim awareness, expansion and synthesis, adversarial checking, external calibration, claim positioning, misreading or misclassification capture, and output verification and revision. The framework is not intended as a universal rule for all AI use, but as a proportional safeguard for high-cognitive-risk contexts in which AI outputs may shape belief, judgment, institutional language, professional responsibility, or consequential action. The paper concludes that responsible generative AI use is not a choice between trust and distrust, but a disciplined relation.