Study: Disclosing AI Chatbot's Persuasive Intent Reduces Persuasion

A new study on arXiv (2608.11794) looks into how being transparent affects the persuasive power of AI chatbots. In this preregistered experiment, 1,500 adults from the UK chatted briefly with a chatbot about one of 60 different policy topics. Everyone interacted with the same chatbot, but they were split into three groups: one with no disclosure (control), one that knew they were talking to an AI (T1), and another that also learned about the chatbot's persuasive goals (T2). The control group showed a 12.6-point attitude shift, while T1 had a 13.1-point shift, similar to the control. However, T2 actually diminished the chatbot's persuasive effectiveness. This study emphasizes the need for transparency about AI's role and intent for better regulation and content clarity.