Central Bank Communication, Conversational AI, and Macroeconomic Expectations
How to cite
Pin, Wei; Ash, Elliott; Song, Lena; Weber, Michael; Xie, Shihan. 2026. "Central Bank Communication, Conversational AI, and Macroeconomic Expectations." AEA RCT Registry. AEARCTR-0018842. https://doi.org/10.1257/rct.18842-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
In this project, we examine whether AI chatbot–based monetary policy information treatments shift households' macroeconomic expectations and improve economic reasoning about interest-rate transmission channels. We conduct an online survey experiment on Prolific, randomly assigning approximately 3,000 U.S. participants to receive either a hawkish or dovish narrative regarding the monetary policy orientation of new Federal Reserve Chair Kevin Warsh, delivered through an AI chatbot interface, with participants randomized to either a minimal single-exchange interaction (control) or a structured two-phase conversational protocol (treatment) powered by Claude Sonnet 4.5. We will evaluate treatment effects on macroeconomic expectations and policy reasoning elicited through survey responses.
Experimental design
We will recruit around 3,000 participants who completed prior waves of our recurring weekly survey from Prolific.
The survey begins with a screener and informed consent, followed by baseline measures of knowledge on economics and the Fed, attitudes towards AI, and attention checks.
We cross-randomizes the narrative framing of Kevin Warsh's monetary policy orientation and the delivery mode in a 2X2 between-subject design. Participants receive either a hawkish or dovish narrative regarding the monetary policy orientation of new Federal Reserve Chair Kevin Warsh, delivered through an AI chatbot interface, with participants randomized to either a minimal single-exchange interaction (Active Control) or a structured two-phase conversational protocol (Treatment). That is, the participants are randomized into four groups: (i) Active Control + Hawkish Narrative (ii) Active Control + Dovish Narrative (iii) Treatment + Hawkish Narrative (iv) Treatment + Dovish Narrative.
After exposure, participants complete a set of post-treatment measures.
Design details
Cited by
No citation data available for this registration yet.