AI-Assisted Oversight? Generative AI and Written Parliamentary Questions

Written parliamentary questions (WPQs) are one of the most frequently used instruments of legislative oversight. Generative AI sharply lowers the cost of producing them, with potential consequences for both the quantity and the quality of oversight. Yet we know little about whether, how and by whom these tools are used in parliaments, or how they reshape oversight itself. This paper addresses both sets of questions in European parliaments (currently Sweden and the United Kingdom). First, we map AI use by applying Pangram, a commercial AI-text detector, to all WPQs from before the release of ChatGPT until today. We document a sharp rise in AI-flagged questions from 2024 onwards and decompose it across parties and legislator characteristics. Second, we examine whether AI changes the nature of WPQs: do early adopters ask more questions, and do AI-flagged questions differ in technical language, topical focus and harshness of tone? We compare MPs who adopt AI with those who do not, and the same MPs before and after adoption. The findings speak to current debates about how established democratic institutions adapt to generative AI, and how these tools may reshape political communication, representation and accountability.