Artificial intelligence-enabled green innovation: Evidence from patent data

Artificial intelligence (AI) is reshaping how knowledge is created and combined, yet raising a sustainability paradox. While AI systems are resource-intensive, they may indirectly contribute to climate mitigation by supporting the development and diffusion of sustainability-oriented technologies. We examine this second-order channel by assessing whether green patents incorporating AI are more technologically influential than those that do not. Drawing on OECD REGPAT patent data from 1977 to 2024, we assemble a large dataset of climate-mitigation inventions, flag AI-enabled patents, and link them to standardized quality indicators and applicant information. We operationalize technological quality using three complementary indicators: originality, radicalness, and forward citations within five and seven-year windows. We estimate Poisson pseudo-maximum likelihood models with high-dimensional fixed effects for citation outcomes and fractional logit models for originality and radicalness, including extensive patent, team, and applicant-level controls. We further implement propensity score matching and reweighting strategies. Across specifications, AI-enabled green patents are consistently associated with higher forward citations and higher radicalness, while no association emerges for originality. We further show that the citation premium is strongest for narrower-scope inventions, declining as patent scope increases, suggesting that the influence of AI-enabled green inventions is contingent upon technological breadth. Overall, AI-enabled green inventions appear to diffuse more widely and to be more radical, but not because they draw on a broader knowledge base. These findings inform debates on AI’s sustainability implications beyond its immediate environmental footprint.