Generative AI for complex urban planning: Pathways, potentials, and challenges

Managing urban complexity has become a defining challenge of the twenty-first century, requiring approaches that transcend conventional planning tools. Generative artificial intelligence (GenAI) has emerged as a promising response, reshaping knowledge production, spatial design, and planning practices. Yet research remains fragmented, with limited understanding of technological evolution and uneven adoption across planning domains. This study systematically reviews 183 publications from 2016 to July 2025, combining bibliometric analysis with thematic synthesis to trace GenAI's evolution, regional distribution, and applications in urban planning. A GPT-based assistant was used to support data organization and coding. The findings reveal that: (1) GenAI has advanced from statistical models and machine learning to GAN- and VAE-based methods, and more recently to diffusion models, large language models (LLMs), and multimodal systems, marking a shift toward high-fidelity, controllable, cross-modal generation; (2) application intensity is highly uneven across the globe, and application directions are differentiated, with Asia emphasizing GAN-based methods while Europe shows stronger adoption of LLMs; (3) urban design, morphology generation, and environmental modeling are relatively mature domains, whereas transportation, participatory planning, and especially land-use planning remain fragmented and exploratory; and (4) research is constrained by narrow datasets, fragmented workflows, and the absence of evaluation frameworks aligned with planning objectives. By synthesizing dispersed applications into a coherent framework, this review clarifies GenAI's methodological progression in urban contexts and outlines integration pathways. It argues that GenAI's transformative potential lies in system-level integration, enabling comprehensive, multi-agent-driven support to meet the challenges of complex urban planning. • Systematic review of 183 studies on GenAI in urban planning (2016–2025). • Employs a customized LLM-based agent for large-scale literature processing. • Traces evolution from statistical models to GANs, VAEs, diffusion, and LLMs. • Reveals global disparities in application intensity and technological orientation. • Calls for system-level integration to meet challenges of complex urban planning.