The purpose of this study is to investigate the ability of Large Language Models (LLMs) to recognize linguistic markers of emotions in Russian and to assess their potential for profiling emotional users of AI. We conducted a series of experiments to verify psycholinguistic models of emotions, including J. Russell’s circumplex model and R. Plutchik’s wheel of emotions. We created synthetic personas with different emotional states and used them to experiment with LLMs, such as YandexGPT 5 Pro, using personalized emotional prompts. To assess the consistency of LLM responses, we employed ANOVA procedures, which allowed us to test hypotheses about differences in how synthetic personas reacted to the same emotional stimuli. The results of our study demonstrate that LLMs and humans structure emotions differently. The novelty of this work lies in the application of a personalized approach to analyzing LLM emotional perception, which allows us to take into account the sociodemographic characteristics and emotional state of communicants. This research is significant because it aims to develop methods for assessing LLM emotional intelligence and creating the foundation for improving emotional AI systems that can provide empathic support in dialogue.