AI-assisted hermeneutics: validating a large language model for the rhetorical analysis of the book of revelation

Abstract This article introduces and validates a framework for the computational analysis of rhetorically complex texts, moving beyond generic sentiment analysis to enable nuanced, large-scale hermeneutic inquiry. As a case study, we analyse the Greek text of Revelation, using the Gemini-2.5-Pro LLM to quantify its rhetoric against a bespoke, theologically-grounded eight-vector lexicon. The framework’s validity is established through strong correlation (mean Pearson’s r = 0.82) with a reliable human-expert baseline (Fleiss’ κ = 0.825), confirming its capacity to map the text’s salient features. The analysis provides quantifiable evidence for key exegetical claims—including the liturgical framing of divine judgment and a 12:1 deception-to-violence ratio for the Beast’s threat—and shows how AI-human divergences can map a text’s implicit theology. This study offers a reproducible and validated heuristic for mapping complex foundational texts, demonstrating a critically-aware partnership between computational analysis and expert exegesis.