AI talk–action gaps and ESG rating sensitivity: Evidence from Chinese A-share firms

Research background:Environmental, social, and governance (ESG) ratings have become increasingly important in investment decisions and corporate evaluation, yet they often rely heavily on publicly available firm disclosures. At the same time, artificial intelligence (AI)related communication has expanded rapidly among listed firms, creating a new setting in which strategic narratives may shape external assessments before real innovation outcomes become fully observable. This raises an important question about whether AI-related disclosure is associated with ESG ratings beyond firms’ verifiable technological progress. Purpose of the article: This article examines whether AI-related strategic disclosure is associated with higher ESG ratings through a gap between firms’ AI-related narratives and their observable AI innovation. It also investigates whether any rating gains associated with such a gap reflect credible forward-looking information or instead capture rating sensitivity to soft disclosure. By linking AI-related disclosure to subsequent stock returns, future operating performance, long-run abnormal returns, and debt financing outcomes, the study seeks to distinguish between informational and distortionary interpretations. Methods: Using panel data for Chinese A-share listed firms from 2010 to 2023, the study constructs a residual-based AI talk–action gap measure that captures AI-related textual disclosure beyond the level predicted by verifiable AI patenting activity and firm characteristics. Firm fixed-effects models are employed to estimate the relationship between this gap and ESG ratings, followed by a set of forward-looking tests on market and operating outcomes. The empirical analysis is supplemented by alternative variable definitions, placebo tests, and cross-sectional analyses to assess robustness and heterogeneity. Findings & value added: The results show that a larger AI talk–action gap is associated with significantly higher ESG ratings, suggesting that ESG evaluators respond not only to observable AI innovation but also to AI-related strategic disclosure in excess of such innovation. The gap is further associated with more positive short-run stock market reactions, but it does not predict stronger future profitability, sustained long-run abnormal returns, or lower debt financing costs. These findings indicate that AI-related soft disclosure is associated with higher ESG ratings and temporarily more favorable equity market perceptions, but its informational content is only weakly validated by subsequent fundamentals. The article contributes to the literature on ESG rating efficiency, strategic corporate disclosure, and sustainable finance in the context of digital transformation.