How AI and Humans Express Comfort Differently: A Corpus-Based Appraisal Analysis

Grounded in the appraisal framework, this study investigates differences in Judgement and Appreciation in the construal of interpersonal affective responses. A 194,141-word corpus was compiled from Reddit user texts and ChatGPT outputs across eight counseling themes. The analysis combined natural language processing with manual annotation. Results show that GPT employed more attitudinal resources but with less dimensional diversity, primarily offering positive evaluations to affirm behaviors and events. Meanwhile, human-authored texts contained fewer resources overall but displayed greater diversity, frequently using negative moral Judgement to signal stance and reduce social distance. In addition, GPT’s negative evaluations were concentrated primarily in event Appreciation, which expanded its expressive range while offering a safer channel for negativity by targeting situational complexity rather than individuals. These findings reveal that both humans and AI are able to mobilize most institutionalized attitudinal resources to construct interpersonal affective responses, yet they diverge in their tendencies: GPT predominantly creates a safe, universally supportive environment through positive evaluations, whereas Reddit users rely on critical moral Judgement to assert stance and enhance emotional resonance. This study advances understanding of how interpersonal affective responses are constructed in human-AI interaction and provides linguistic evidence to inform the development of AI-mediated communication.