AI Shopping Agents: What Walmart's Rules Mean for Your Site

AI shopping agents are AI assistants that find, compare and buy things for a person, and this month Walmart wrote rules for them. On October 21, 2026, Walmart said agents may buy only with the customer's approval or under a spending cap, and it blocks agents that break its terms. An agent visiting your website moves on when a price, stock level or return policy is missing.

Key facts

• Walmart, October 21, 2026: customers can give each AI agent its own profile, with approval on every order or a spending cap such as $50, $100 or $200. Agents that break the terms get blocked.
• Google, October 15, 2026: Universal Cart went live in US Search and the Gemini app. The merchant stays the seller of record.
• Sierra and Meta, October 6, 2026: the Personal Agent Protocol was announced as an open standard for how personal agents deal with businesses. The v0.1 spec was still due later in October.
• PYMNTS, October 23, 2026: in a survey of 2,191 US consumers, 40.2% would send an agent to a large marketplace and 16.1% to a brand's own website.
• Canada: Universal Cart is US only, Walmart's release names no Canadian rollout, and the protocol isn't a product yet. Details below.

What is an AI shopping agent?

An AI shopping agent is software that shops on a person's behalf: it searches, compares options, fills a cart and, with permission, pays. The person gives the goal ("a rain jacket under $150, delivered by Friday"), and the agent does the clicking.

Some agents run inside an AI app, such as ChatGPT or Gemini. Others run inside a browser and load web pages the way you do. Bret Taylor and Clay Bavor of Sierra put the current state plainly in their October 6 announcement: "Today most personal agents use websites and apps the way people do."

For a business owner, the risk is quiet. An agent that can't find your price, can't read your return policy, or gets stuck on your contact form doesn't complain. It moves to the next site on its list.

We covered the traffic side in the new traffic class: agents acting on behalf of users.

What changed this week for AI shopping agents?

Three things moved in October: Walmart set terms, Google shipped a cart, and a group led by Sierra and Meta proposed a standard. OpenAI, which launched in-chat checkout in 2025, has stepped back from it.

Walmart set the rules (October 21). Walmart's release describes three pieces:

1. Agent Profiles, due within the next month. Each agent a customer authorizes gets its own profile in their account, so its purchases show up separately. The customer can require approval for every purchase or set a spending limit, and can delete the profile at any time.
2. Walmart Agent Gateway, a faster route for agents built on MCP (Model Context Protocol, a standard way for AI tools to plug into other software). It starts as a beta for companies and developers. ActuIA's report describes it as invitation-only, with identity checks, rate limits and category restrictions.
3. Terms for agents. Agents can't act beyond what the shopper asked for, never receive card numbers, and can't add or change payment methods. Walmart's returns process applies "including if an agent doesn't act as intended." Agents that don't follow the terms will be blocked from the site.

Walmart also says agents see the same prices a person sees. Its line: "We price products, not people."

Google's Universal Cart went live (October 15). In Google's launch post, Suresh Ganapathy, Senior Director of Product Management for Consumer Shopping, describes a cart that holds products from several retailers across Search (including AI Mode) and the Gemini app. It watches for price drops and restocks. Shoppers pay with Google Pay or move the items to the store's own cart. On who owns the sale, Ganapathy is direct: merchants "will remain the seller of record." Checkout runs on the Universal Commerce Protocol, which we explain in our UCP explainer. It is rolling out in the US.

Sierra and Meta proposed the Personal Agent Protocol (October 6). The protocol covers sign-in and gives companies visibility into what agents do on their website, their APIs or their own customer-service agent. Partners named: Genesys, Instinct, Rocket, Shopify, Stripe and Walmart. The principle is two-sided: consumers choose what access their agent gets, and "companies set parameters for what those agents can do." The spec was not out at announcement.

OpenAI stepped back from in-chat checkout. ChatGPT's Instant Checkout launched in September 2025. On March 24, 2026, TechCrunch reported that OpenAI was deprioritizing it to focus on product discovery, letting merchants use their own checkout. OpenAI's statement: "the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide."

ChatGPT and Google both now let buyers finish on the merchant's own checkout. The agent does the research; your site still closes the sale.

How do AI shopping agents find and read a business website?

Agents reach a business three ways: by loading its web pages like a person, by pulling a structured feed or catalogue, or by calling a tool the business publishes. Most small businesses today are only reachable the first way.

The first route is the one you control this week. An agent reading a page has the same problems a rushed human does, without the patience. If a price only appears after a chat pop-up, or the hours are inside an image, the agent may not find them. We wrote about the third route in WebMCP: the standard your site needs and in our primer on the Model Context Protocol.

Being found in the first place is a separate job. Agents start where AI search starts: an answer from ChatGPT, Gemini or AI Mode that names a short list of businesses. That is AEO (answer engine optimization, meaning getting named inside ChatGPT and Google AI answers). An agent can't buy from a business it never heard of.

Which retailers let AI shopping agents in, and which block them?

The large retailers have split. Walmart lets agents in on its terms, Google built a cart that merchants join, and Amazon and eBay block outside agents.

Amazon's position, from its statement on the lawsuit as reported by Retail Dive, is that third-party shopping agents "should operate openly and respect service provider decisions whether or not to participate." According to ActuIA, a hearing on Perplexity's motion to dismiss is set for November 20, 2026.

Across the wider market, blocking is the minority choice. A DOSS study of 364 readable retailer robots.txt files, collected June 17, 2026, found about 1 in 10 major retailers block AI crawlers. Marketplaces blocked at about 23% and apparel at about 21%. The study checked robots.txt only, not checkout. If you're unsure what your own file says, start with what robots.txt actually does in 2026.

We looked for QVC's stance and found no published statement, so it's not in the table.

What do AI shopping agents mean for Canadian businesses?

For Canadian businesses, the October launches are mostly a preview. Universal Cart is rolling out in the US, Walmart's rules cover shopping with Walmart and Sam's Club, and Google's Canadian onboarding is next year.

What applies in Canada today, from the sources:

• Google Universal Cart: not announced for Canada. Google's launch post says it is rolling out in the U.S. and names no other country.
• UCP in Merchant Center: Google's onboarding help page says the integration "is gradually rolling out in the U.S., with Australia and Canada to follow next year." In May, Google had said Canada would come "in the coming months," per its May post.
• Walmart's agent rules: the release covers Walmart and Sam's Club and doesn't mention Walmart Canada or any country.
• Personal Agent Protocol: a proposed standard, not a launched product, so it applies nowhere yet.
• Agents that browse: these work anywhere. A Toronto shopper who asks an agent to compare three local stores gets a result based on what those three websites say.

The protocols are US-first. Browsing agents aren't tied to a country, so your website's text is the part of agentic shopping that already reaches Canadian customers.

What about service businesses, not just retailers?

Service businesses get agent visits too, just for different jobs: comparing providers, checking prices and hours, and filling a booking or quote form. Sierra's Personal Agent Protocol announcement talks about companies in general, including their websites, APIs and customer-service agents, not only stores.

An accountant, a clinic or a contractor won't plug into Universal Cart. But a person can already ask an agent to "find three bookkeepers near Liberty Village who list their prices and can start this month." The agent then reads each site. The businesses that state prices in plain text, show availability and have a short, working form are the ones it can act on.

The PYMNTS data shows why being easy matters. In its October 23 report, shoppers chose where to send an agent mostly for practical reasons: 36.6% cited easier or more reliable returns, 34.9% better selection. Translate that to services: a clear guarantee, a full list of what you offer, and a visible price range.

What should you change on your website now?

Make the facts an agent needs readable as plain text on the page, check that you aren't blocking agents by accident, and keep your forms short. None of this needs a protocol.

For the structured data step, our guides cover LocalBusiness schema and Organization schema. Boutique shops can see how we approach retail sites on our boutique retail page.

If you want to know where your site stands, our free site check looks at how your pages read to search engines and AI.

What are the limits of AI shopping agents right now?

Most of what launched this month is early. Walmart's Agent Profiles are due within the next month, the gateway is a beta, and the Personal Agent Protocol had no spec at launch. ChatGPT's in-chat checkout was scaled back because, per TechCrunch's source, users "weren't using the chatbot to actually help them make purchases."

The survey numbers point the same way. PYMNTS found 57.9% of AI shoppers usually still buy on Amazon even when AI helps them find the product. Agents are changing how people research. Fewer of them are finishing the purchase on their own yet.

So fix the basics before buying an integration. Every change in the table above helps human visitors too.