ABSTRACT The growing capabilities of large language models (LLMs) pose important questions for rhetorical theory and pedagogy. This article offers an overview of how LLMs like GPT work and a consideration of whether they should be considered rhetorical agents. To answer this question, the article considers structural and argumentative similarities in classical theorizations of rhetoric and the philosophy of Wilfrid Sellars. GPT’s particular method of encoding statistical patterns in language gives it some rudimentary semantics and reliably generates acceptable natural language output, so it should be considered to have a degree of rhetorical agency. But it is also badly limited by its restriction to written text, and an analysis of its interface shows that much of its rhetorical savvy is caused by the highly restricted rhetorical situation created by the ChatGPT interface.