Abstract
Religious language continues to permeate contemporary discourse, even inostensibly secular domains such as environmental activism and climate changedebates. This paper investigates how explicit and implicit forms of religiouslanguage appear in climate-related texts produced by secular and religiousnongovernmental organizations (NGOs). We introduce a dual methodologicalapproach: a rule-based model using a hierarchical tree of religious termsderived from ecotheology literature, and large language models (LLMs) operatingin a zero-shot setting. Using a dataset of more than 880,000 sentences, wecompare how these methods detect religious language and analyze points ofagreement and divergence. The results show that the rule-based methodconsistently labels more sentences as religious than LLMs. These findingshighlight not only the methodological challenges of computationally detectingreligious language but also the broader tension over whether religious languageshould be defined by vocabulary alone or by contextual meaning. This studycontributes to digital methods in religious studies by demonstrating both thepotential and the limitations of approaches for analyzing how the sacredpersists in climate discourse.