This paper outlines a theoretical framework to develop methodological processes for integrating Artificial Intelligence (AI) into urban planning. The proposed approach emphasizes the importance of establishing a solid methodological foundation before any operational implementation in professional practice. Given AI’s disruptive and rapidly expanding presence across all domains of human activity, there is a growing tendency to adopt AI tools directly, often overlooking the risks associated with such uncritical usage. In some areas of the international debate on AI and urban planning, this trend is evident in the immediate development of techniques for applying Large Language Models (LLMs) without grounding them in a coherent theoretical structure. This framework should instead account for the diverse potentials of LLMs across various stages of spatial planning, including: the analysis and construction of knowledge bases, the interpretation of territorial phenomena, quantitative modeling of urban systems, the anticipation of evolutionary trajectories, and the elaboration of future planning scenarios. While some instances of “direct application” can be found in current literature, this study deliberately follows a different path—advocating for a more structured and critically informed integration of AI into planning theory and practice.