Geocoding is a fundamental process for epidemiology, as it allows for the spatial analysis of disease distribution, such as dengue, with accuracy being a critical factor for study validity. The objective of this study was to test three geocoding methods using a database of probable dengue cases from 2007 to 2024 in the Federal District, Brazil. The techniques evaluated were: the Geocoding API (Google Maps), the Nominatim API (OpenStreetMap - OSM), and a method developed with Artificial Intelligence (AI) that uses a Large Language Model (LLM) with Retrieval Augmented Generation (RAG) based on the CNEFE - IBGE address database. In the results, the Geocoding API (Google Maps) obtained the highest return rate (99.6%), followed by the LLM (AI) (98%) and Nominatim (OSM) with only 11.5%. However, in terms of accuracy, the Google Maps API showed the largest errors and outliers. The OSM technique was the most accurate in the validated sample, but it extrapolated results to other countries in cases of high ambiguity. The LLM (AI) approach proved robust, both due to its high return capability and its precision errors. Furthermore, it offers greater flexibility in identifying and correcting errors in the geocoding process. The OSM and LLM (AI) techniques are viable candidates, and the combination of methods can optimize the final quality in future analyses and ensure the applicability of spatial analyses in public health contexts.