The Power of Conversation: A Survey Experiment on Ai-Generated Information and Political Behaviour

Citizens are increasingly using generative AI platforms such as ChatGPT to seek political information. Despite concerns about the reliability of these platforms, extensive usage by citizens of these tools is poorly understood, including implications for political knowledge and voter behaviour. Crucially, these platforms differ from other digital technologies such as news websites, social media, and search engines, in their human-like interactivity and capacity to summarise and synthesise information on demand. These unique characteristics have the potential to lower the barrier to accessing political information, thereby increasing citizens' perceptions of political knowledge and efficacy. Yet, there is little causal research on the relationship between generative AI chatbots and political behaviour. We field a preregistered survey experiment involving a controlled information seeking task, holding substantive content constant to test whether AI’s unique features increase individuals' internal political efficacy, as well as two potential mechanisms: its effect on factual knowledge and trust in information. We focus on the context of the European Parliament, characterised by low baseline knowledge and trust, high complexity, and weak partisanship.