The rehabilitation of abandoned buildings presents numerous advantages, which are widely discussed in the literature. However, it requires information about each building, given the specificity of each example. Without a unified database, information on buildings with potential for rehabilitation is scattered across articles (journalistic), papers, theses, and dissertations, making access difficult. This work aims to investigate the use of artificial intelligence as a tool to identify these buildings in Brazilian territory. The work used a tool based on neural networks that operates through textual vectorization, transforming unstructured data into analytical structures that AI systems can interpret. Initially, 32,678 documents on the subject were collected on a global scale, of which 1561 were considered relevant to the study. From the sample of 1561 documents, 723 buildings were identified in Brazil, of which 138 were identified as having potential for rehabilitation, reaching 82 with complete adherence to the characteristics sought. The results are presented through a visual interface, showing the identified examples on a map with the generation of a database of documents, photos, and information about the identified buildings.