LLM-Agents for Geo-Planning: Semi-Automated Pipelines for Prompts, Rules, and Quality Control (GeoAI-Territory, Part II)

AbstractThe rapid growth of spatial data and the increasing complexity of urban and regional planning create a pressing demand for scalable, cost-efficient, and transparent decision-support tools. Large Language Models (LLMs) have recently shown promise as intelligent agents capable of interpreting heterogeneous geospatial datasets, generating adaptive workflows, and supporting human decision-makers. This paper introduces the concept of LLM-Agents for Geo-Planning, focusing on semi-automated pipelines that integrate prompts, rule-based reasoning, and multi-layer quality control. Unlike traditional GIS automation, LLM-agents can flexibly adapt to different regulatory frameworks and local contexts, from European INSPIRE directives and SDG-oriented policies to rapidly expanding African urban environments and large-scale land monitoring systems in China. We present a unified pipeline architecture where domain-specific prompts guide task execution, explicit rules ensure compliance with local planning standards, and human-in-the-loop validation guarantees accuracy and accountability. Three case studies demonstrate the versatility of the approach: (1) land-use zoning and digital twins in European municipalities, (2) ecological restoration planning in African peri-urban regions, and (3) large-scale agricultural land assessment using satellite data in China. The results highlight how LLM-agents can reduce analysis time by 5–10×, improve reproducibility, and make advanced spatial planning accessible to regions with limited technical resources. This work contributes a globally relevant framework for integrating LLMs into geospatial governance, urban planning, and sustainable territorial development.