Emotional Differences Between Human- and AI-Generated Texts

Abstract AI is often conceptualized primarily as a cognitive tool, with its outputs assumed to convey information rather than emotional expression. This study challenges that assumption by systematically comparing the emotional characteristics of human- and AI-generated text across 25 languages using five emotional measures: sentiment polarity, sentiment categories, Valence-Arousal-Dominance (VAD), basic emotion classification, and fine-grained emotion classification. The results reveal a systematic difference in the emotional profiles of AI- and human-generated text across languages. Compared with human writing, AI-generated text generally exhibits higher valence, arousal, and dominance, while expressing fewer negative emotions. These findings extend our understanding of the affective characteristics of generative AI and highlight important implications for applications in customer service, marketing, personalized communication, and AI-generated content detection.