Testing an AI Chatbot in a Citizens’ Assembly: Design, Use, and Effects on Deliberation

In the context of the increasing interest in artificial intelligence (AI) technology, practitioners of mini publics have expressed a growing interest in incorporating AI solutions into mini publics. Given the high costs associated with mini publics, the primary interest of practitioners is to reduce costs and enhance workflow efficiency. Conversely, technology and AI could be employed within the deliberation process itself, with distinct objectives and in different phases. I expand the classification proposed by McKinney (2025) upon by the differentiating between the phases of process planning and management (1), random selection of participants (2), learning (3), deliberation (4), facilitation and decision making (5), and follow up (6). In an interdisciplinary research project at the University of Wuppertal, scholars from the departments of political science and computer science have developed an AI-chatbot intended for deployment in phases 3 and 4 of a citizens assembly. This paper offers insights into the first mini public in Germany, wherein participants were given the opportunity to utilize an AI system that was specifically designed for this particular use case. The AI chatbot was introduced and tested in a mini public on the topic on how to foster democratic participation within a mid-sized municipality in North Rhine-Westphalia. Participants were encouraged to utilize the AI system to pose questions concerning the information they had received from experts during the course of the process. The database was filled with the information provided by the experts, and the participants were able to access it via a question-and-answer chatbot system whenever they liked. A mixed-methods approach was employed to evaluate the use of AI and its impact on the participants. This approach included participatory observation throughout the entire process, as well as a quantitative preliminary, interim, and follow-up survey and qualitative interviews to provide deeper insights into the process. The objective of this study is to assess the implementation of the AI system within the mini public and its subsequent impact on the participants' deliberation process while addressing the following research questions: • What were the participants' reactions to the AI utilization within the mini public? • What impact did it have on the mini public and the deliberation process? • Was there value added and what does this first experiment implies for following mini publics where AI might be used? The study offers answers derived from the data gathered and critically discusses the results against the backdrop of deliberative democracy theory and the risks of AI utilization in general, such as bias, disinformation and hallucination.