Consensus or Chaos? The Impact of Listed Companies' Generative AI Disclosure on ESG Rating Divergence: A Signaling Theory Perspective

Publicly listed companies are increasingly disclosing information related to generative artificial intelligence (GAI), yet it remains unclear how environmental, social, and governance (ESG) rating agencies—as key information intermediaries—interpret these complex signals. This study extends signaling theory to examine the effect of corporate GAI disclosures on ESG rating divergence, highlighting that the impact depends on the interaction between signal complexity and signal credibility. Using textual analysis of annual reports from listed firms and combining these with ESG ratings from multiple mainstream agencies, we conduct an empirical investigation. The results show that: (1) low-complexity disclosures, such as merely mentioning GAI concepts or products, have no significant effect on rating divergence; (2) disclosures describing concrete applications and use cases of GAI effectively reduce information asymmetry and thus significantly converge rating divergence; (3) high-complexity disclosures, such as claims of possessing underlying technologies, without sufficient credibility support, increase interpretive difficulty and thereby amplify rating divergence in an inverted U-shaped pattern; and (4) firms' ownership of AI-related patents, serving as a key credibility signal, can substantially mitigate the negative effects of high-complexity disclosures. This study reveals the mechanisms through which emerging technology disclosures shape ESG rating divergence, offering a novel explanatory dimension for understanding its sources, and providing important practical implications for corporate disclosure strategies and market regulation.