Walmart says its returns process covers an agent's mistakes. Decide what your store says before an agent orders from you
When Walmart published rules for AI shopping agents on October 21, one line stood out: its returns process applies even when an agent doesn't act as intended. Most online stores haven't decided what happens when an agent buys the wrong thing. Here are four rules you could pick, and how to choose.

Walmart's new terms for AI shopping agents, announced October 21, 2026, say its returns process applies even if an agent doesn't act as intended, purchases need approval by default, and agents never see card details. Smaller online stores should pick their own rule before agent orders arrive: treat agent mistakes like any return, accept them only with approval on record, add limits for certain products, or block agents. For most shops, the first two are the sensible starting points.
What Walmart decided
On October 21, Walmart announced how it will handle AI agents shopping on behalf of its customers. There are three parts. Agent Profiles, launching within a month, give each agent its own profile inside a customer's account, so its purchases show up separately, and let the customer require approval for every purchase or set a spending limit such as $50, $100 or $200. The Agent Gateway, starting as a beta for companies and developers, gives agents a direct way to shop over MCP, the protocol many AI tools use to connect to services. And a set of terms says what agents can and can't do.
Most of the terms are what you'd expect. Agents don't receive card details and can't change payment methods. They only get the information the task needs. Purchases need the customer's approval by default, and Walmart may ask for more approvals even within a spending limit. Agents that don't follow the rules get blocked, according to TechCrunch.
Walmart also says agents see the same prices a person would see, with offers and fees shown before purchase, and that customers don't need a Walmart+ membership to use an agent.
One line is different, because it answers a question most stores haven't asked yet: "Walmart's returns process applies – including if an agent doesn't act as intended."
Why that line matters to a smaller store
Walmart has the scale to absorb returns. Your store may not. But the question it answered is coming for everyone who sells online: when an AI agent orders the wrong size, the wrong color or ten of something instead of one, what happens?
Right now, at most stores, nobody has decided. The order looks like any other order. The customer contacts support and says "my assistant bought this, not me." Support makes a call on the spot, and next week a different person makes a different call. The question isn't whether an agent will buy the wrong thing. It's who pays for the trip back.
Agent orders are still a small share of most stores' sales. That's exactly why now is the cheap time to pick a rule: before it's a pattern, and before a frustrated customer posts about it.
The question isn't whether an agent will buy the wrong thing. It's who pays for the trip back.
Four rules you could pick
Rule 1: An agent's mistake is a normal return. This is Walmart's choice. Your existing return window and conditions apply, whether a person or their agent placed the order. It's simple, it's friendly and support doesn't have to investigate who clicked. The cost is that you absorb returns caused by software you don't control.
Rule 2: Normal returns, if the order was approved. If your checkout or the agent platform records that the customer approved the purchase, it's treated like any order. If there's no approval on record, you still accept the return but you might handle it differently, for example as store credit. This rewards customers who use approval settings and gives you something concrete to point to.
Rule 3: Normal returns, with limits on certain products. Some products are expensive to take back: made-to-order items, perishables, large furniture, anything opened. You keep Rule 1 for most of the catalog and set clearer conditions for those, the same ones that apply to people. The key is that the conditions are written down and shown before purchase, so an agent, and its user, can see them.
Rule 4: No agent orders. Some sites have chosen to block agents entirely; TechCrunch reported that Amazon has begun doing so. That's a legitimate decision for a store whose products really need a person to choose. It means turning away some customers, and it only works if you can actually detect and stop agents.
Decision guide
Four return rules for agent orders
| Consideration | Rule | Trade-off |
|---|---|---|
| Normal return | Same window and conditions for everyone. | You absorb agent mistakes. |
| Approval on record | Normal returns when the customer approved. | Needs order-source data. |
| Product limits | Clear conditions for costly items. | Product data must carry them. |
| No agents | Block agent orders. | Turns away some customers. |
How to choose
Start with your current return rate and what a return costs you, shipping both ways plus handling. If returns are cheap and rare, Rule 1 is easy and makes you the kind of store people trust their assistant with.
If returns are expensive, Rule 2 or 3 is usually better. Rule 2 depends on knowing whether an order was approved, so ask your commerce platform what it records about agent orders today. Some will show the source of an order; many won't yet.
Rule 4 is the right choice less often than it looks. Most of the reasons stores consider it, wrong items and surprise orders, are better handled by clear return conditions than by a wall.
Whichever you pick, it has to be in writing on your returns page, in the same plain language as the rest of it, so an agent reading the page can tell its user.
Making the rule real
Tag the order. If your platform can tell you an order came from an agent, keep that information on the order. You'll want to know how often agent orders come back, and why.
Give support one sentence. "We treat orders placed by an assistant the same as any other order" or whatever your rule is, written in the support playbook.
Put product limits in the product data. If Rule 3 applies, make the return condition a field on the product, not a line hidden in a policy page. Agents read structured information more reliably than paragraphs.
Keep prices and fees the same. Walmart's statement that agents see the same prices as people is worth copying. If an agent sees a different total than a person would, you'll spend more time on disputes than you save.
Decide what counts as "not as intended." Wrong size, wrong quantity and wrong item are easy. An order the customer simply regrets is a normal return under your usual rules, not a special case because an agent placed it.
Check after a quarter. Look at agent orders, returns and complaints. If the numbers are small and the rule is working, leave it. If one product category keeps coming back, tighten that category, not the whole store.
If you'd like help making your store ready for agent orders, from how products are described to how returns are recorded, that's what our ecommerce team does. Talk to us.
Related work: Spin Market.
How the work moves
From rule to practice
- Pick the rule
Based on return cost and rate.
- Publish it
Plain words on the returns page.
- Tag agent orders
Keep the source on the order.
- Brief support
One sentence in the playbook.
- Review in a quarter
Adjust by category, not store-wide.
We build alongside the people doing the work.
Bring us the problem you’re thinking about.