Talk or action? Unveiling the nature and depth of climate disclosures in Islamic banks using machine learning

Climate risk has become a board-level and supervisory priority, yet little is known about how Islamic banks—subject to both conventional regulation and Shariah principles—communicate their climate commitments. Drawing on 838 annual reports from 103 Islamic banks across 25 jurisdictions during 2015–2024, this study develops a four-step natural-language–processing pipeline that combines rule-based precision with context-aware machine learning. Findings reveal that climate language appears in 84 % of bank-year observations but remains governance-heavy and metric-light, strategy narratives account for 64 % of TCFD-aligned sentences, whereas metrics & targets comprise only 11 %. Mentions of ESG governance and climate stress-testing surged six-fold after 2020, reflecting alignment with NGFS supervisory guidance and intensified regulatory uptake in Southeast Asia and the Gulf. Nonetheless, renewable-finance and net-zero narratives remain peripheral (< 3 %). These results highlight a disclosure regime characterized by rapid narrative diffusion yet limited quantitative substance, challenging assumptions about Islamic finance’s presumed ethical advantage and informing regulators seeking more decision-useful climate reporting.