<p>Many scientists suspect that AI can replace humans in most situations, while others think that AI cannot intuitively replace humans. Emoticons, punctuation-based symbols representing emotions, serve as the focus of this study which investigates the abilities of ChatGPT, Claude, and DeepSeek to identify and interpret facial emotions conveyed through symbolic emoticons. A total of 34 emoticons, ranging from common to rare punctuation combinations, were presented to each model using the prompt: “What do you think:) is?”, without additional context. Model responses were evaluated based on recognition accuracy, emotional interpretation, and explanation coherence. The results showed that all three models successfully identified the majority of emoticons as face expressions, with DeepSeek demonstrating the highest accuracy, followed by ChatGPT, and Claude. Distinct interpretive styles were observed: ChatGPT gave concise, function-oriented responses; Claude offered culturally rich interpretations; and DeepSeek adopted a socially and psychologically analytical tone. Our findings suggest that while language models can recognize faces and basic emotions comparably to humans, they show limitations in detecting nuanced emotional states. We conclude that AI may contribute meaningfully to the study of emotional intelligence and its applications in sentiment analysis, human–computer interaction, and digital communication.</p>