Study: AI chatbot intent disclosure halves persuasion effect

A new preprint from Adrian Rauchfleisch and Andreas Jungherr, posted on arXiv, tackles a core question in AI governance: does disclosing that you're talking to an AI actually reduce its persuasive power? In a preregistered experiment with 1,500 UK adults, participants held short conversations with an identical persuasive chatbot about 60 policy issues. Some saw no disclosure, some saw a prominent label saying they were interacting with an AI, and a third group saw that label plus an explicit statement of the chatbot's persuasive intent and instructions.

The results are striking. In the control group, the chatbot shifted attitudes by 12.6 points on a 100-point scale. Adding the AI-identity disclosure did essentially nothing—attitudes shifted by 13.1 points. However, when the chatbot's persuasive intent was also revealed, the effect dropped to 6.3 points—roughly half. Intent disclosure also made participants view the campaign's methods as less acceptable and increased support for stronger penalties against it. The authors argue that current regulatory emphasis on what a system is (AI vs. human) misses the real lever: what the system is trying to do. This suggests transparency mandates for chatbots should go beyond provenance labels and require disclosure of persuasive goals.