Early experiments on the persuasive power of AI chat systems raised big hopes in political communication: what could be better than letting AI do the hard work of persuading people? It’s not that simple. Yes, experimental studies show that AI-generated content can convey information and even change opinions. But these studies usually measure persuasion under forced-exposure conditions. They do not show how people feel about being approached by an AI with persuasive intent. Findings from a new study with @OuzhouAdi suggest that AI-mediated political outreach does not necessarily help campaigns and may carry negative consequences. In a preregistered 2x2 experiment in the US and UK (N = 1,800 per country), we test how people react to announced political contact: either from a human campaign volunteer or an AI-mediated system, and in each case with either an informational or persuasive purpose. We find two clear patterns: 🔸 Persuasion penalty: outreach with explicit persuasive intent is rated more negatively than purely informational outreach 🔸 AI penalty: AI-mediated outreach is rated more negatively than human outreach across almost all outcome variables, in both countries Importantly, the negative effects of AI-mediated outreach are especially pronounced when the contact is framed as informational. In other words, the costs of AI appear even in contexts that may initially seem more legitimate or less problematic. The negative reaction also extends beyond the contact itself to the organization behind it, with implications for trust, avoidance, and reputation. Much of the existing research asks: can AI change opinions? We ask: how legitimate and acceptable does AI-mediated political outreach appear to people in the first place? If we focus only on persuasive effectiveness, we risk overlooking the larger costs of deploying such systems. Working Paper: arxiv.org/abs/2603.27413v1