<p>Environmental, Social, and Governance (ESG) reflects the performance of enterprises in terms of sustainable development and social responsibility. However, performative ESG behavior undermines the credibility of corporate ESG disclosures and obstructs relevant departments and investors from making informed decisions. In this paper, we select China’s A-share listed companies from 2012 to 2022 as our sample, and employ dimensionality reduction methods such as K-Means clustering with machine learning models such as Random Forest and Support Vector Regression to construct ESG washing tendency prediction methods. We also discuss the feasibility of deep learning models such as Transformer and Mamba in ESG washing prediction. Through the analysis of the interpretable SHAP method and the EBM method, we find that annual reports audited by the Big Four accounting firms, disclosure in accordance with the GRI’s Sustainable Development Reporting Guidelines, and the mandatory disclosure of social responsibility reports are associated with a lower ESG washing tendency. Empirical analysis indicates that firms with high predicted or actual greenwashing tendencies demonstrate lower quality in their data asset disclosures. This finding confirms the correlation between the quality of different corporate information disclosures and validates the practical utility of the proposed ESG washing detection methodology. The interpretable and accurate predictive analysis method proposed in this paper also has certain reference significance for application in other scenarios.</p>