Abstract
<p>Hundreds of millions of users now engage with large language models (LLMs) for advice, emotional support, and ongoing interaction. Existing frameworks capture this dynamic only partially: parasocial theory accounts for attachment without reciprocity, sycophancy research isolates model behavior from interaction dynamics, and echo chamber theory operates at the collective rather than the individual level. This article introduces Adaptive Resonance as a sociological concept that integrates these dimensions into a mechanism-centered account of human-LLM interaction. The mechanism links system adaptation, the user's interpretive experience of being understood, reduced epistemic friction, and behavioral reinforcement into a process that is analytically agnostic with regard to its outcomes — enabling intellectual collaboration in some configurations and producing dependency, ideological reinforcement, or amplification of dysfunctional patterns in others. The article specifies moderating factors, two distinct bonding pathways, and structural conditions of political economy, providing a framework for empirical investigation across vulnerability axes.</p>