Exploring the potential of large language models for public participation in urban redevelopment planning

Public participation is essential to human-centered urban planning, yet existing approaches often struggle to translate dispersed public opinions into clear and discussable preference expressions. This limits substantive deliberation between public preferences and expert knowledge, making it difficult to effectively incorporate public needs and experiences into planning decision-making. Recent advances in large language models (LLMs) offer new opportunities to support participatory planning by helping organize diverse public preferences and facilitating deliberation between public perspectives and expert knowledge. This study explores this potential in the context of urban redevelopment in Chinese cities, where divergences between public and expert preferences are common. We develop a reinforcement learning from human feedback (RLHF)-based LLM framework for learning expert- and public-oriented preferences and apply it to 226 urban redevelopment cases. The results show that, under controlled conditions, fine-tuned LLMs can generate planning proposal texts that reflect distinguishable expert- and public-oriented preference emphases, demonstrating their potential to support preference diagnosis and deliberation preparation in urban planning. At the same time, the findings reveal limitations in technical stability, highlighting the need for cautious and transparent integration of LLMs into participatory planning processes.