From cart to card: what shoppers expect from agentic commerce?

Consumers are ready to let AI do their shopping discovery. Most still want a say before it pays.

In Riskified’s September 2026 Agentic Commerce Pulse survey, 57 percent of 5,100 consumers across five markets said they’d let AI compare products or prices for their holiday shopping. Just seven percent would let it buy gifts without asking first.

That comfort is growing. Half of shoppers are more at ease now with an agent completing a purchase without their immediate approval than they were in our last survey six months ago. New personal agents are adding momentum: Meta’s Muse passed 2.5 million downloads in its first two weeks, and Instinct, still invite-only, is nearing $1 billion in annual transaction volume.

For merchants, this shift raises practical questions. When an agent places an order, who’s the customer behind it? And if it orders incorrectly, who handles the refund?

Here’s how far shoppers will let AI go today, and how to prepare your shop for what comes next.

Trust thins out as the agent gets closer to the transaction

Eight in 10 consumers have used AI for at least part of their shopping in the last six months. Most of this use happens early, as shoppers research and compare prices. Use drops at each step closer to payment, and just 11 percent have purchased within an AI tool.

AI use falls from 56% to 10% as consumers approach the purchase

Holiday plans follow the same pattern. More than two-thirds of shoppers (68 percent) said they are likely to use AI at some stage of their holiday shopping, and most would let it find gift ideas or compare products and prices.

So what holds shoppers back at the final step of purchase?

The top three consumer concerns about AI-powered shopping all involve money changing hands:

• Unauthorized purchases made by an AI agent
• Fraudsters accessing their AI assistant or saved payment information
• AI choosing the wrong product

A weak product recommendation costs a shopper a few minutes. An incorrect or unauthorized purchase costs real money and requires a refund request. 

Shoppers want conditional autonomy

When asked which approach consumers would be most comfortable using for purchases made by an AI shopping agent, nearly three in four consumers (74 percent) said that they would allow an AI agent to make purchases in some form. Few would give it free rein:

• 43 percent want to approve each purchase upfront
• 18 percent would let the agent buy within rules they set and ask for approval on exceptions
• 10 percent would let the agent buy within their rules and notify them afterward
• 3 percent would let the agent buy without any limits or approvals

Call it conditional autonomy: shoppers will give AI more authority, as long as they define the limits. Asked what would make them comfortable, almost half (47 percent) named the ability to approve purchases before they happen. About four in 10 named each of these: notifications before a purchase is finalized, the ability to cancel or reverse a purchase, and spending limits they set.

Shoppers also want transparency. More than two-thirds (68 percent) said they’d always want to know when they’re interacting with an AI agent instead of a human. That visibility matters to merchants too, well past checkout. Customer service teams handling returns, refunds, and support chats will need to know whether they’re talking to the shopper or the shopper’s agent.

Where shoppers stand differs by market

Every market shows the same drop in comfort from research to purchase. What changes is how far along shoppers are in various markets.

How shoppers used AI in the past six months

Shoppers in Mexico and Brazil use AI more at nearly every stage, and about twice as many have already bought through an AI tool. In the US, UK, and Japan, four in 10 or more still wouldn’t let an agent buy anything for them.

Who’s responsible is still an open question

Shoppers haven’t settled on who should run their agents or fix their agents’ mistakes.

If an agent is tricked into buying a counterfeit or misreads instructions, 51 percent of consumers expect the AI platform to issue the refund. Yet Muse and Instinct both state in their terms that users are responsible for their agents’ actions. In practice, those disputes are likely to reach the merchant.

What this means for merchants

As more shoppers try personal agents like Muse and Instinct, agent-led activity will likely concentrate where trust is thinnest today: checkout, payment, and everything that follows. Two shifts stand out.

Merchants need to determine the customer behind the agent. Approval steps, notifications, and spending limits help shoppers feel in control. These controls don’t confirm whether an order comes from a real customer using a legitimate agent, a hijacked account, or malicious automation. That’s harder to judge when Muse and Instinct browse websites without identifying themselves and appear as ordinary human visitors.

“The agent did it” will likely become a new refund claim. Some of these claims will reflect genuine agent errors. Others will be fraudulent claims, and merchants will need a way to tell the two apart.

To help merchants do that, Riskified is extending its identity risk intelligence to assess the person behind the agent, for the order and the service interactions that follow.

Want to see how agentic commerce could affect your approvals and disputes? Talk to an expert today.