Shoppers are happy to let AI help them shop, but very few will let it spend their money on its own. In a new Riskified survey, only 7 percent said they are comfortable with AI buying something without their approval. Just 11 percent have actually completed a purchase inside an AI tool.
What the survey found
Riskified, a company that helps online stores fight fraud, polled more than 5,100 consumers worldwide for its Agentic Commerce Pulse report. Agentic commerce means an AI agent shops or buys for you. Agents named in the report include Muse and Instinct.
- 80 percent have used AI for some part of shopping.
- 74 percent would allow agent purchases in some form.
- 68 percent are likely to use AI for holiday shopping.
- Half are more comfortable than six months ago with an agent buying without immediate approval.
The gap shows up at checkout. Comfort drops the closer AI gets to spending money. The report describes shoppers as embracing AI as an assistant faster than as an autonomous buyer.
Control and trust
Many shoppers want to keep a hand on the wheel. 43 percent still want to approve every purchase. 29 percent would let an agent buy automatically within rules they set, with approval for exceptions or notices after the fact.
Trust varies by who runs the agent. One-third would trust a general AI platform such as ChatGPT or Gemini. 17 percent would trust a retailer or brand they already shop with. Only 9 percent would trust a bank, card company or digital wallet. Shoppers also named payment security and AI picking the wrong product as top worries.
Who pays when AI gets it wrong
This is the sharpest finding. 51 percent said the AI platform should issue refunds after a mistake. But the report says the terms and conditions of both Muse and Instinct make users responsible for what their agents do.
Jeff Otto, Riskified's chief marketing officer, said shoppers have not delegated the blame. In practice, he said, the dispute lands on the merchant. He also warned that agents could become a new way to abuse refunds, and that stores will need to tell a real customer with a legitimate agent from a hijacked account or malicious automation.
What to do
Anyone trying an AI shopping agent can check the terms first to see who is responsible for errors. Setting limits on what it may buy, and asking for approval, keeps the risk smaller. Merchants may want to plan for refund disputes tied to agent purchases.
What is not known yet
The material does not say whether any platform plans to change its terms. Riskified sells fraud tools to merchants, so its findings come from a company with a stake in the issue.
The takeawayBefore letting an AI agent buy for you, read who is liable for its mistakes, because the terms may put the cost on you.