Large Language Models (LLMs) have revolutionized text generation tasks, offering significant potential in supporting Environmental, Social, and Governance (ESG) report generation by enhancing efficiency, consistency, and data analysis. However, existing LLM-based systems often suffer from usability challenges, such as unintuitive interfaces, content authenticity concerns, and limited interaction mechanisms. To address these issues, this paper proposes an innovative interface design for LLM-driven ESG report generation systems, incorporating three key features: real-time output display to improve usability through mental model alignment; categorizing and visually highlight AI-generated content to enhance trustworthiness and reduce review workload; and an LLM-based assistant agent that facilitates user interaction by explanations of the AI-generated content, information retrieval, and customized report revision. A user testing demonstrates significant improvements in usability, reduced cognitive load, and enhanced user satisfaction compared to existing systems. By bridging the gap between algorithmic capabilities and user-centered design, this study contributes to advancing the usability and reliability of AI-assisted content generation systems with a cost-effective and flexible approach.