Digital tools or digital facades? institutional conditions for credible AI-enabled carbon regulation in China's multi-level governance

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

As China digitizes carbon governance, the central challenge is shifting from information scarcity to the credibility of digitally generated evidence. This paper examines the institutional conditions under which AI-assisted screening and auditing can strengthen verification and enforcement in a multilevel governance system, as well as how strategic compliance may weaken these effects. It conducts a structured qualitative comparison of Zhejiang, Hebei, and Inner Mongolia, combining cross-case analysis with process tracing based on regulations, government reports, platform documentation, enforcement materials, and investigated cases. Regulatory credibility is evaluated along three dimensions: information integrity, measurement validity, and enforcement closure. The findings show that digital systems have improved data integration, reporting frequency, record traceability, information retrieval, and anomaly screening. However, publicly available evidence is considerably stronger for information acquisition and screening than for the final resolution of individual cases. One official investigation provides direct evidence of reporting-time manipulation and record fabrication, while indicator substitution remains an institutional vulnerability rather than a documented practice. The available evidence does not substantiate platform capture. The study therefore neither estimates the independent causal effect of AI nor ranks the three provinces by regulatory credibility. Instead, it argues that AI-enabled regulation becomes credible only when technical outputs are connected to evidence beyond the control of reporting entities, traceable revision histories, valid indicators, independent verification, and documented final dispositions. The paper concludes by recommending auditable data-provenance standards, institutional separation between algorithmic detection and verification, and standardized public reporting on enforcement closure.