Evaluating the Impact of Report Readability on ESG Scores: A Generative Ai Approach

Abstract

This study explores the relationship between the readability of sustainability reports and ESG scores for U.S. companies using GPT-4, a generative AI tool. The findings reveal a positive correlation between context-dependent readability scores and the average of multiple ESG scores, whereas their standard deviations exhibit a negative correlation. Conversely, existing text-dependent readability scores reflecting word features show no correlation with ESG scores. Moreover, we observe a correlation between readability and ESG scores among companies with lower social visibility, where transparent disclosure is essential for accurate ESG evaluation. These results point to the usefulness of context-dependent readability in ESG evaluations. In particular, it suggests that the stability of ESG evaluations is related to the high level of readability that takes context into account.

Keywords: ESG ratings, readability, Sustainability, GPT, AI