The purpose of the Article is to demonstrate and illustrate the importance of specific descriptions of the subdialects vs. the generalization of facts in dialect atlases. A method is used to compare individual examples in different of the dialect Atlas parts and individual subdialect materials using publications, records, and conversations with subdialect speakers over several years. In comparison, the deep latgalic subdialect Ziemers from the High Latvian dialect is used where the Malenian variant is spoken. The results convince that dialect atlas materials do not show the full state of any particular subdialect these days. The sociolinguistic nature of certain subdialects and their applications, the impact of different subdialects or dialects, and the relative phenomena in different dialects of languages can only be specifically known in individual subdialect studies and descriptions of their results, necessitating a nationally funded large research programme, which could also form part of a joint programme of dialect studies in neighbouring countries, including expeditions, dialectologists' discussions and joint publications. Artificial intelligence is a tool that can already help in various structuring and classification processes in dialectology as well. It is evolving as it becomes increasingly useful, yet, as its current resources show, the tool should be used responsibly and academically fairly, making sure the data is reliable. The development of artificial intelligence is and will depend heavily on the availability of the results of our past research.