As there is currently no dedicated text classification for French queries in the context of diabetes management, this research addresses that gap by developing a solution. This study explores the use of advanced language models for emergency case detection in diabetes management. We focus on fine-tuning three French language models CamemBERT, FlauBERT, and RoBERTa using a dataset of 2,140 questions generated from medical sources and ChatGPT. The dataset, labeled manually with the assistance of healthcare professionals, includes both emergency and non-emergency questions with varying lengths. We evaluate the models based on performance metrics such as accuracy, precision, recall, and F1-score. Our results indicate that CamemBERT outperforms the other models in detecting emergency cases with 0.989 in accuracy. This work not only demonstrates the efficacy of CamemBERT for this specific task but also lays the groundwork for its integration into a comprehensive chatbot system for diabetes management. Future efforts will focus on enhancing this system to provide improved support and intervention for patients.