Animal emotion detection has emerged as a critical domain in understanding animal welfare and enhancing human–animal interactions. This study evaluates the performance of GPT-4, a generative AI model, in recognizing and classifying pet emotions from images, with a particular focus on dogs. The research was conducted in two phases: a general pet emotion classification across multiple species and a dog-specific classification. In Phase 1, GPT-4 achieved an overall accuracy of 50.2%, reflecting its baseline capability in handling diverse animal images. In Phase 2, accuracy significantly improved to 76.7% due to refined prompts and the use of a targeted dataset. Sentiment analysis of the model’s textual justifications revealed alignment with visual cues in correct predictions and highlighted challenges in ambiguous cases. These findings demonstrate the potential of generative AI in animal emotion detection and emphasize the importance of targeted datasets and advanced prompt engineering. This study contributes to bridging the gap between AI capabilities and practical applications in animal welfare and behavioral research.