Abstract
<title>Abstract</title> <p>Corporate engagement with the United Nations Sustainable Development Goals (SDGs) has grown rapidly, but the quality of SDG reporting has not kept pace. Most corporate SDG claims remain symbolic — generic references without measurable targets or verifiable evidence. Existing assessment methods tend to be either rigorous or scalable, but not both. This study addresses this gap by developing an agentic artificial intelligence framework that assesses the credibility, rather than merely the presence, of corporate SDG claims. The framework maps each claim to the 17 SDG goals and 169 targets, and scores it on a five-dimensional credibility rubric, including Specificity, Measurability, Temporality, Evidence, and Consistency (SMTEC), grounded in legitimacy, signalling, and accountability theory. We apply the framework to 30 sustainability reports from companies listed on the Stock Exchange of Thailand for the 2024 reporting year, scoring 6,325 SDG-mapped claims. Validation against two domain experts yields a three-rater Krippendorff’s ordinal alpha of 0.744; the framework’s agreement with each expert is statistically indistinguishable from the agreement between the experts themselves. Across the corpus, 8.7% of claims are symbolic, 68.6% partially substantive, and 22.8% substantive. Notably, firms tend to under-claim rather than over-claim their SDG engagement, declaring far fewer goals than their reports substantively address. The framework offers regulators, investors, and standard-setters a scalable, theory-grounded instrument for distinguishing substantive from symbolic sustainability disclosure.</p>