Comfort is ahead of usage
Riskified published results on 27 October from a September survey of 5,100 adults in the United States, United Kingdom, Japan, Mexico and Brazil who had recently shopped online. Eighty percent said they had used AI somewhere in the shopping process, but only 11% had completed a purchase inside an AI tool. Forty-three percent wanted to approve every purchase, while 29% would allow automatic buying within rules they set. Because the research was commissioned by a fraud-and-risk vendor, merchants should treat the figures as directional rather than universal.
Liability expectations do not align
The survey highlights a governance problem for automated commerce. Fifty-one percent of respondents said the AI platform should issue a refund when an agent makes a mistake, even though platform terms may place responsibility on the user. Sixty-eight percent wanted to know whenever they were interacting with an AI agent. For merchants, unclear identity and authority could translate into disputes, refund abuse and customer-service friction, especially if an agent exceeds a shopper’s stated limit or selects the wrong item.
What it means for merchants
Define how agent identity, delegated authority, purchase limits and refunds are recorded before accepting autonomous orders, and preserve evidence that can be used in disputes without exposing unnecessary customer data.
Read the original announcement
Riskified · 27 Oct 2026This brief summarizes the linked primary source. Claims and reported figures have not been independently audited by Payment Guide.