In the context of China’s carbon peaking and carbon neutrality goals and high-quality growth strategy, corporate ESG performance has attracted considerable attention. As regulatory requirements for ESG disclosures tighten in China’s capital markets, firms face the challenge of effectively implementing ESG strategies and accurately reporting ESG data. Meanwhile, AI has emerged as a key tool for addressing ESG challenges. However, most existing studies focus on technological innovation and overlook AI’s integration with the stakeholder theory, especially its role in corporate decision-making and stakeholder engagement related to ESG goals.

This paper combines the stakeholder theory and cost-benefit analysis to construct a multi-period dynamic game model, and systematically analyzes how AI affects corporate ESG performance. Using data on AI patents and ESG-related news from A-share listed companies from 2011 to 2021, it empirically tests the impact of AI on ESG performance and examines the role of media sentiment in this process. The results demonstrate that AI significantly improves ESG performance, with a stronger effect in larger firms, non-polluting industries, and regions with a higher level of digital economic development. Mechanism testing reveals that AI affects ESG performance primarily by promoting ESG investment and enhancing corporate transparency. Additionally, media ESG sentiment plays a crucial moderating role: Positive sentiment amplifies AI’s positive effect on ESG performance, while negative sentiment reverses this impact, with neutral sentiment showing no significant effect.

This paper has the following contributions: First, it offers a fresh perspective on AI’s role in improving corporate ESG performance, providing valuable insights into AI research in the context of sustainability. Second, it develops a theoretical multi-period dynamic game model based on corporate-stakeholder interactions, integrating media ESG sentiment and cost-benefit equilibrium to explore optimal ESG decision-making under different conditions. Third, it refines AI measurement by utilizing patent text data and digital economy industry classifications, and introduces media sentiment as an external governance force, enriching the corporate governance theory.