Large language models are everywhere, from chatbots and search engines to content generators and decision-making systems. But what is actually happening when they 'speak'?

Discourse Machines argues that LLMs are not neutral tools but active producers of meaning, knowledge, and power. Drawing on critical theory, post-structuralism, and discourse analysis, Simon Lindgren situates LLMs within longer traditions of thinking about language, power, and knowledge. Through close readings of interactions with ChatGPT, Claude, Gemini, and other systems, he reveals how these machines reproduce truth, perform authority without understanding, simulate care without embodiment, and enact the 'death of the author' while projecting a coherent voice. As these systems become embedded in education, healthcare, journalism, and governance, the stakes could not be higher.

Offering a critical perspective on generative AI, Discourse Machines is essential reading for understanding what happens when language itself is automated, and what we must do to reclaim discourse from the machines.

Simon Lindgren is Professor of Sociology at Umeå University.