AI-Enabled governance for sustainable cities: evaluating the credibility of urban climate disclosures and greenwashing risks

Achieving the Sustainable Development Goals (SDGs), particularly SDG 11 (sustainable cities), SDG 13 (climate action), and SDG 16 (accountable institutions), requires governance that is transparent and trustworthy. However, the credibility of city sustainability disclosures varies, and the increasing prevalence of greenwashing makes it difficult to distinguish genuine commitments from symbolic claims. This study develops an AI-enabled analytical framework that integrates large language models (LLMs) and explainable AI (XAI) to evaluate 537 city responses from the Carbon Disclosure Project (CDP) Cities dataset. Grounded in institutional theory and the dynamic capabilities perspective, the framework conceptualizes sustainability disclosures as both institutional signals and manifestations of governance capability. The analysis follows a three-stage process combining LLM-based scoring, interpretability analysis, and external validation. SHAP analysis of the surrogate XGBoost model (R 2 = 0.995) suggests that terms related to governance, assessment, monitoring, and citizen engagement are associated with higher LLM-derived scores. Chi-square analysis further revealed heterogeneous greenwashing patterns, characterized by symbolic overpromising in A-List cities and structural or procedural weaknesses in Non-A-List cities. Overall, the findings suggest that institutional pressures shape how sustainability commitments are communicated, while governance capabilities influence the extent to which these commitments are supported by implementation mechanisms. The proposed framework provides a scalable and interpretable approach for evaluating the credibility of urban sustainability disclosures and supporting more accountable climate governance.