Seeing the Positives: How Humans and Large Language Models Interpret Emotional Framing in ESG Disclosures

Firms increasingly use positive language in environmental, social, and governance (ESG) disclosures, yet little is known about how such framing shapes investor judgment or whether artificial intelligence interprets these narratives similarly. Drawing on psycholinguistic theory, we combine eye-tracking with a 2 × 2 experiment manipulating emotional framing (positive vs. neutral) and temporal framing (retrospective vs. prospective). Positively framed retrospective disclosures increase investors’ willingness to invest and reduce perceived risk. Eye-tracking evidence shows that positive language attracts greater visual attention and induces more regressive eye movements, indicating heightened cognitive scrutiny that partially offsets its persuasive effects. We replicate the experimental design using eight state-of-the-art large language models (LLMs). The models exhibit systematic, human-like sensitivity to both emotional and temporal cues. Methodologically, the study provides one of the first parallel validations of human and LLM-based interpretations, revealing shared mechanisms in ESG narrative processing. The findings inform ESG disclosure regulation, sustainability communication, and AI-assisted financial decision-making, while advancing interdisciplinary research on emotional framing in non-financial narratives.