AI Overstatement: How Shopping Agent Claims Overstate Capabilities and Undermine Service Transparency

This project accompanies the manuscript “AI Overstatement: How Shopping Agent Claims Overstate Capabilities and Undermine Service Transparency”. The study introduces the construct of AI overstatement as a property of individual claims about AI shopping agents and examines how specific message features contribute to the inflation of perceived capabilities. The dataset consists of consumer-facing web pages from three AI shopping agents (ChatGPT Shopping, Rufus, Daydream). From these pages, 421 claims where extracted; among them 152 claims about the agents’ capabilities where considered for this research. Each claim is coded for intensity, evidence support, verifiability, disclosure of limitations, and anthropomorphism, and assessed through a verification-based overstatement judgment.