The contribution is devoted to the application of the BERT language model for the analysis of drug data in order to identify side effects described in the instruction texts. The work focuses on the problem of text data extraction using named entity extraction methods, which allows for more accurate identification and classification of side effects. A drug database is used as an example, which demonstrates the practical significance of the study. The stages of data preparation and preprocessing required for model training are described in detail. The authors consider the adaptation of BERT for the analysis of medical instructions, which enables the effective identification of drug side effects. The study includes the analysis of the text description of instructions and the identification of similarities in the side effects of various drugs. The BERT-based model is capable of rapidly processing and analyzing large volumes of text information, which significantly simplifies the process of extracting data on side effects from instructions. The results of the work emphasize the potential of using modern language models in the field of medicine and pharmacology, opening up new horizons for the automation of medical text analysis.