Retailers are picking sides on AI shopping agents. Your brand is on every side at once.

When a reporter at The Verge asked Meta's Muse agent to find the cheapest granola bars per piece, it came back with options from Sam's Club, Walmart and Costco. It also offered a caveat: Amazon might have had the lowest price, but Muse was hard-blocked from browsing it, so it couldn't check. That detail comes from Mia Sato's October 22 report on what happens when agents try to shop, and it's the clearest picture yet of a split that matters to every consumer brand.

Granola bars are a CPG product. Whoever makes them sells at all four of those retailers. And for that one shopper, on that one day, the brand's Amazon listing simply didn't exist.

Three positions, one week

Retailers' views on AI agents have been forming all year. In the third week of October they got much easier to see.

Retailer or platform	Position	What they've said or done
Amazon	Closed to outside agents	Sued Perplexity over its agent. Asked Meta to remove Amazon from Muse. Runs its own assistant.
Walmart	Open, with written rules	Agent terms, Agent Profiles with spending limits, and an MCP-based Agent Gateway in beta (Oct 21).
BigCommerce stores	Open by default	WebMCP turned on for stores with an opt-out; checkout stays on the merchant's storefront (Oct 21).
Target	Open to partners, guarded	Partnerships with OpenAI, Google and Microsoft; safeguards against unauthorized agents.
The RealReal	Open, guarded	Lets outside agents browse and log in, with safeguards against disruptive automation.

Sources: The Verge, Walmart, Commerce (BigCommerce), TechCrunch.

Amazon's position is the simplest to state. Spokesperson Maxine Tagay told The Verge the company "requested that Meta remove Amazon" from Muse, and that "third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate." Amazon is not against agents; its own assistant can monitor prices and reorder items. It's against agents it doesn't run.

Walmart went the other way, with conditions. Its October 21 announcement set out terms that read like a contract for robots. Customers must approve agent purchases by default. Agents don't receive payment card details. Agents can access only the information the task needs. "Agents that do not follow these terms will be blocked from Walmart's site." And on pricing: "We price products, not people," meaning an agent sees the same price as a human.

BigCommerce, the ecommerce platform, took the most open position. With Cloudflare, it turned on WebMCP, a way for browser agents to use a store's functions directly, across its merchants by default, with an opt-out.

The question for a brand isn't which side is right. It's which shelf an agent shopper will actually see.

Why it's not settled by the courts

You might expect a lawsuit to decide this. So far the courts have left it to the retailers. In August, the Ninth Circuit vacated the preliminary injunction Amazon had won against Perplexity's shopping agent, reasoning that the user, not Perplexity, was the one accessing Amazon. The same opinion noted Amazon can still set its own rules through its terms of service.

In practice, that means the rules will be written one retailer at a time, in terms of service and bot settings, not by a judge. Expect them to keep changing.

How big is agent shopping, really?

Not big yet, but no longer hypothetical. On Meta's earnings call, Mark Zuckerberg said about 25% of Muse users use it to shop, with an average purchase around $200, out of roughly 10 million sign-ups in the US and Canada, according to Payments Dive. That is one agent, launched in September.

The experience is still rough. Sato found Muse was caught by Target's bot check one day and let through the next, and that it suggested a West Coast grocery chain that didn't ship to her New York address. Target's spokesperson, Loni Monroe, said success can depend on how an agent goes about a task, and that some parts of the site may be reachable when others aren't.

So the honest summary is: a small but growing number of shoppers are handing purchases to agents, the agents work inconsistently, and where they can't go, they go somewhere else.

What this means for a CPG brand

Most consumer brands don't control any of these decisions. You don't decide whether Amazon lets Muse in. What you do control is what an agent finds at the retailers that let it in. Three things follow.

The open shelves get relatively more important

If an agent shopper can reach Walmart, Target and Costco but not Amazon, your Walmart, Target and Costco listings are the whole shelf for that shopper. Many CPG brands put most of their online effort into Amazon because that's where most of their online sales are. For agent shoppers, that effort is invisible. Tip 30 covers why retailer listings feed AI answers in the first place; the agent split makes that doubly true.

This doesn't mean shifting budget away from Amazon. Amazon's own assistant, which we covered in our Alexa for Shopping playbook, is likely to reach far more shoppers than outside agents this year. It means the second and third retailer can't be an afterthought anymore.

Agents compare, so contradictions cost more

A human shopper who sees 12 count on one site and 10 count on another might shrug. An agent asked for the best price per piece will do the math, and if your data is inconsistent it will either get it wrong or skip you. Walmart's "same price as a human" rule helps here; it means what the agent sees is your real shelf price. The facts around that price, size, count, ingredients and claims, are on you. Tip 5 is the long version.

Your own site may become a store for agents

If you sell direct on BigCommerce, agents may already be able to shop your store through WebMCP unless you've opted out. On Shopify, OpenAI says merchants are already integrated with ChatGPT shopping, so your products may be showing up there whether or not you applied. Whether to welcome that is a real decision with tradeoffs on fraud, returns and margin, and tip 33 lays them out. For most CPG brands, direct is a small channel, and the bigger question is still the retail shelf.

The repeat-purchase angle

One detail in Walmart's announcement deserves more attention from CPG brands than it got. Walmart says its Agent Gateway will help agents shop for a customer's "usuals" without sharing the customer's personal information or purchase history with the agent. Its example: if a customer tends to buy sourdough bread or a particular brand of milk, Walmart will help the agent find those items.

That is a CPG feature. Food, drink, household and personal care products are bought on repeat, and a lot of brand loyalty is really habit. If agents start reordering through a retailer's own memory of "the usual," incumbents gain. The brand already in the basket stays there, and the agent has little reason to look around. For a challenger brand, the moment of trial still happens when a person, or an agent asked an open question, compares options. That makes the open-question answers, the "best oat milk for coffee" kind, more valuable than ever, because they're where new products get into the basket in the first place.

Walmart says Agent Profiles launch within the next month and the Gateway starts as a beta for developers, so none of this is at scale yet. It is worth watching through the holidays.

The part nobody can plan for yet

The deeper shift is that the "shelf" now depends on who, or what, is looking. Two shoppers asking for the same product can get different sets of retailers depending on which assistant they use. One in Amazon's search bar sees Amazon. One in Muse sees everything except Amazon. One in ChatGPT sees merchants drawn from OpenAI's mix of direct and scraped product feeds, which we looked at in today's guide to ChatGPT Shopping.

Traditional retail measurement wasn't built for this. Share of shelf assumed one shelf. Now there are several, and they overlap only partly. Companies like XENA Intelligence work on measuring how brands show up in AI answers, and SearchPanel is one tool built for that kind of check. Whatever you use, the useful question to ask about your category is simple: when an agent shops for this, where does it go, and are we there?

It's tempting to wait until the retailers settle. They may not, for years. Amazon has every reason to keep its walls up, and Walmart has every reason to keep them down. A brand on both shelves will be living with the split either way. The work that holds up under every scenario is the boring kind: complete listings, consistent facts, and checking what the machines say.