Machines Referencing the Qurʾan: Citation Patterns in LLM-Generated Islam

Abstract

Large Language Models ( LLM  s) are increasingly capable of generating Islamic religious discourse perceived as meaningful by human recipients. This article examines one dimension of this capacity: the ability of LLM  s to perform istishhād – the citation of Qurʾanic verses to support theological positions – when role-playing six Muslim personas representing distinct positions in contemporary Islamic discourse. Drawing on 480 LLM -generated responses from four models ( GPT -4o, GPT -5, Claude 4 Sonnet, and Gemini 2.5 Flash), accessed via API and web interfaces, the article analyses 2,941 Qurʾanic references, examining citation frequency, verse distribution, and cross-persona similarity. Results show that LLM  s produce theologically differentiated citation patterns broadly consistent with known positions in Islamic discourse. A secondary finding concerns substantial variation in citation density across models and access channels, raising methodological questions about reproducibility. The article concludes that non-human Islam, while mechanistically novel, exhibits structure, patterns, and theological coherence warranting sustained academic attention.