The increasing deployment of Internet of Things (IoT) devices in energy and environmental monitoring provides high-resolution data on household consumption patterns. This paper presents a simulation-based framework that leverages Large Language Models (LLMs) to analyze real smart meter data, together with scientific literature and policy texts, in order to generate structured, actionable recommendations for sustainable energy use. The LLM outputs are encoded in machine-readable formats (e.g., JSON) suitable for ingestion by Smart Contracts within blockchain ecosystems. Policy execution is evaluated ex-post by applying both baseline static rules and LLM-guided policies to real IoT datasets, thereby simulating how household consumption would change if users followed the recommendations. Results indicate that households complying with LLM-guided policies achieve 12-18% higher alignment with renewable generation periods and receive 17% more token rewards compared to static rule baselines. These findings demonstrate the potential of combining IoT data, LLM-driven recommendations, and blockchain-based incentives to support automated, evidence-based policy enforcement, and highlight the opportunities for future real-world deployment in sustainability-oriented energy management.