Abstract
This article argues that “AI use” is too blunt a category for judging what happens when a student writes a humanities paper with a large language model. Taking the most intensive case, a paper built with AI from research to final review, it follows the changing relation between student, model, sources, and draft. In research, the model’s first value is orientation: it makes ignorance navigable by producing excess routes, names, terms, and possible sources that still have to be verified, weighted, rejected, or absorbed into an active corpus. In planning, its failures become usable as a negative helper: generic, overbuilt, or wrongly proportioned outlines give the student something to damage through reading, source-pressure, and scale. In writing and review, the danger is false completion, as fluent paragraphs can make unfinished source-relations appear already organized, while a finished-looking draft can hide relations the student still cannot answer for. The exchange becomes useful only when clichés, compressed claims, smooth transitions, late corrections, and detector-resistant surfaces are made answerable to source-bound work. The article therefore reframes authorship as answerability rather than purity from AI. A strong paper built through sustained work with an LLM may remain real academic work while shifting what the final product proves: less comprehensive prior command of a field than dense command over the corpus activated, tested, and made answerable through the process.