Group decision-making (GDM) is a critical process across various domains, including policy planning, healthcare prioritization, and crisis management, where multiple stakeholders collaborate to evaluate alternatives and strive for shared outcomes. In this context, the fuzzy consensus model (FCM) provides a structured framework for quantifying agreement and iteratively guiding preference alignment under uncertainty. However, this model and its variants typically rely on numerical thresholds and standardized feedback mechanisms that do not adequately capture the cognitive diversity, motivational dynamics, and communicative needs of heterogeneous decision-makers. To address these limitations, this study introduces an integrated architecture that combines the FCM with large language models (LLMs) and Retrieval-Augmented Generation techniques. The proposed approach enhances consensus building by transforming numerical feedback into context-aware, persuasive, and psychologically adaptive guidance. Feedback messages are personalized via psychometric profiling based on the Big Five model. Message generation follows structured prompting strategies grounded in dual-process theories of persuasion. In addition, external knowledge sources are incorporated to improve transparency and epistemic reliability. The system is evaluated through a two-phase experimental protocol. The results show that LLMs can generate persuasive feedback tailored to individual profiles and, when integrated into the FCM, contribute to more efficient consensus processes by reducing the number of iterations and improving alignment.