Discourse Depot

Discourse Depot is a public, independent scholarly project that uses large language models as research instruments to examine how generative AI is described, explained, and naturalized in popular, technical, and academic discourse. Founded in October 2025 by Troy Davis, a librarian at William & Mary Libraries, the project operationalizes critical discourse analysis frameworks as detailed system instructions (often exceeding 4,000 words),constrains their outputs with JSON schemas, and applies them to texts about AI. The premise is deliberately reflexive: the same statistical text-generation mechanisms the project critiques are turned on the discourse surrounding them — using the machine to critique the machine. The central concern is the gap between the eloquence of AI outputs and the mechanics that produce them, and the anthropomorphic language that fills that gap — language that attributes thinking, understanding, and intent to systems that perform probabilistic pattern-completion. By making that language the object of study, Discourse Depot treats AI literacy as a question of reality and accountability: whether responsibility for an engineered system disappears into metaphors of autonomous mind, or concentrates in design, deployment, and the human labor and corporate decisions behind the product. Discourse Depot comprises: A growing corpus of analyses of AI-related texts, produced through several analytical frameworks (Metaphor, Anthropomorphism & Explanation; Political Framing; Critical Forensic; Critical Activist; and others); The analytical instruments themselves — the system-instruction-and-schema designs that generate the analyses, advanced under a "prompt as scholarship" pedagogy in which the prompt is the scholarly artifact; Pedagogical materials, including the Glass Box Syllabus, an alignment of the audit prompts with the ACRL Framework for Information Literacy, and an educator's FAQ on generative AI; Interactive learning objects (LLM Training Journey, LLM Inference Journey, and others) and a companion library of standalone, citable analysis artifacts. The accompanying essay, "Looking Through the Glass Box," sets out the project's argument in full.