Household inflation expectations have become an explicit target of monetary policy, yet

they respond weakly to official communication and sit persistently above professional

forecasts. As people increasingly consult generative-AI chatbots about the economy, we

examine whether such conversations move inflation expectations and what makes them

persuasive. Across three preregistered experiments (N = 3,772), participants forecast

U.S. inflation over the next twelve months, discuss their forecast with a partner, and can

then revise it. An AI partner moves forecasts substantially more than a human partner,

whether or not participants can converse with it (Study 1). Randomizing five features of

the chatbot's conversational style independently, we find that challenging, rather than

affirming, the participant's view drives persuasion (Study 2). A chatbot that combines the

most persuasive features outperforms a generic prompt and leaves participants more confident in

their revised beliefs (Study 3). Conversational AI can thus move expectations that official

communication has struggled to reach.