Artificial Intelligence Adoption and Corporate ESG Performance: Evidence from a Refined Large Language Model

Lihao Shen¹Zhengrong Li²Yongqing Liang³Yiqiang Feng²Zhanyu Zhang^2*

• ¹University of California Los Angeles, Los Angeles, United States
• ²Central University of Finance and Economics, Beijing, China
• ³National University of Singapore, Singapore, Singapore

The convergence of artificial intelligence (AI) as a transformative economic force and the rise of Environmental, Social, and Governance (ESG) criteria as a core corporate objective creates a critical yet empirically ambiguous nexus. A central barrier to understanding whether AI adoption truly improves ESG performance is the authentic measurement of firm-level AI usage from corporate disclosures. We address this gap by advancing the foundational LLM-based framework of Jin et al. (2024); our refined pipeline uses a domain-adapted model (Qwen2.5-72B) and a granular classification scheme that isolates authentic 'Applied' technology from mere corporate discourse. Analyzing Chinese A-share firms from 2009 to 2022, this refined indicator reveals a robust positive association between AI adoption and ESG performance. We trace this effect to two primary channels: the fostering of green innovation and the enhancement of internal control quality. This positive association is concentrated in larger, technology-intensive firms, a finding consistent with theories emphasizing the roles of complementary assets (RBV) and absorptive capacity (TOE). Our study provides credible evidence on when and for whom AI contributes to corporate sustainability and offers a transparent methodology for constructing authentic technology adoption measures from textual data. Our study not only provides credible evidence on the AI-sustainability link but also highlights the adoption challenges that create a "digital ESG divide," offering targeted policy implications for an era of escalating institutional pressure.