How AI Affects Deliberative Processes and Their Outcomes? A Randomized Experiment on AI-Assisted Deliberation

There is a growing literature on the possibilities of AI to improve spaces of democratic deliberation by enhancing representation, argumentation and consensus-building, for example (Summerfield et al., 2025). There is already some evidence that AI-tools can perform similar tasks as the core functions of democratic deliberation, and therefore they seem to have potential in terms of helping groups of citizens to find common ground (Tessler et al., 2024), co-creating new ideas (Poole-Dayan et al., 2025), and representing excluded viewpoints (Fulay & Roy, 2025). Critics emphasize risks of AI, such as the erosion of epistemic autonomy and agency, and sacrificing real human interaction and communication for algorithmic opinion formation (Saura Garcia, 2025). However, there is still relatively little empirical evidence on the impact of the use of AI technologies on deliberative processes and their outcomes. This paper addresses this gap by investigating how the use of AI in a deliberative process affects participants’ knowledge, attitudes, perceived quality of deliberation and evaluations of the quality of deliberative outputs. We analyze the results of a randomized lab-in-the-field experiment organized in Finland in Spring 2026. The topic of deliberation in the experiment is the use of AI in higher education. The experiment includes about 400 university students and it follows 2x2 factorial design. The experimental treatments vary in terms of the use of AI in learning and in the integration of different viewpoints. Based on pre- and post-deliberation survey data, we analyze how different uses of AI affect participants’ knowledge gains, learning intensity, attitude changes and trust. Based on post-deliberation survey data, we explore participants’ perceptions and evaluations of the quality of deliberation and its outcome. Our results shed light on how the use of AI tools affect citizen deliberation as well as participants’ perceptions of the process and its outcomes. Therefore, the paper adds to an on-going scholarly debate on the boundaries of the application of AI technologies in democratic processes.