The Wall Street Journal's "The Shopping Bots Are Coming for Amazon" (Nov. 1) lays out the facts: Meta's Muse agent shops across stores, Amazon blocked it, and Walmart and Shopify signed with Meta instead. On its Q3 call, Amazon said more than 400 million customers used Alexa for Shopping in the last 12 months and that it expects Alexa for Shopping to be the primary shopping agent customers use to shop Amazon.
400M+
customers used Alexa for Shopping in the last 12 months
Source: Amazon Q3 2026 earnings call, Oct. 29, 2026
Agentic commerce is the shift to AI shopping agents doing the shopping on a person's behalf, from the first question through, increasingly, to the purchase. What the past six weeks made concrete is that those agents now reach products on two tracks. The first is the retailer agent: it belongs to one retailer and shops only that retailer's catalog, which is what Alexa for Shopping is on Amazon. The second is the answer engine, the general AI shopping service that shops the open web: ChatGPT, Gemini, Google AI Mode and Perplexity, and now Meta's Muse, the ones Amazon calls horizontal agents. They find products wherever they are listed online, on a brand's own site, in a Shopify store, or at any retailer that lets them in. Both decide what to include and what to recommend from product data, but the retailer agent reads the retailer's listing and the answer engine reads your catalog. Amazon just shut its door to the second kind. Walmart opened its door to them. Your product data has to work on both sides.
Amazon shut the door: its own agents shop its own catalog, and outside agents are blocked. Walmart and Shopify opened theirs. Every other retailer is choosing between those two postures, and the choice decides which product data job it creates for the brands and retailers selling through it. Amazon's position is the clearest. Alexa for Shopping is the agent for Amazon's own catalog, U.S. customers who use it spend 50% more per order, and outside agents are not welcome: Amazon blocked Muse shortly after its launch, citing safety concerns, and has blocked Grok Bot as well. At the same time, Shop Direct now lists more than 250 million products from other stores and Buy For Me lets customers purchase some of them without leaving Amazon, so the catalog Amazon's agents shop keeps getting larger. For a brand or retailer, the consequence is that Amazon has to be worked on its own terms: shoppers and Amazon's agents both decide from the Amazon listing, and no amount of readiness on the open web reaches them.
On the answer engine track, Muse reached more than 10 million agents set up in its first six weeks with shopping as the top use case, and a growing list of retailers have let it in, including Walmart, eligible Shopify merchants, Best Buy, Sephora, Wayfair, Ulta Beauty and Dick's Sporting Goods. Walmart is the nuance worth noting: it has a retailer agent of its own and is also building an Agent Gateway so a shopper's outside agent can transact with it under limits the shopper sets, which puts Walmart on both tracks at once. The other answer engines were already here: ChatGPT, Gemini, Google AI Mode, Perplexity and Claude read the same open product data Muse does, and have been shopping it for longer.
Amazon also said it anticipates working with horizontal agents, so the two tracks will overlap at the edges. The split itself holds. As more retailers ship agents of their own, the question for each one is whether it operates like Amazon, keeping its agent to its own catalog, or like ChatGPT, shopping the open web.
Because it shows where the buying decision is moving. The Journal's account of the threat to Amazon's advertising business describes a mechanism: AI shopping agents evaluate product details and prices to decide what to buy, rather than taking in advertising that sways a person scanning a results page, and if the agent is choosing, the sponsored slot has less to act on. Amazon says agentic shopping from outside agents is not meaningfully affecting Amazon Ads today, and that is probably accurate, but it describes the current state rather than the direction.
The direction is the one Agentic Commerce Optimization (ACO) is built around. Agentic shopping engines interpret product data directly and determine which products qualify before a shopper ever reaches a search results page. The product card they assemble is, in Retailgentic's framing, the unit of competition, and the decision moves from a paid placement in front of a person to a product card an agent builds from attributes. That holds on both tracks: Alexa for Shopping assembles its picks from the Amazon listing, and answer engines assemble theirs from your product page and feed. We covered where advertising goes from here when we launched AdGentic. The other side of that trade is that product data becomes the main determinant of whether a product is included at all, ahead of page design and ad spend.
ACO is the process of preparing a product catalog for AI-powered shopping: making every product ready to be found, understood and chosen by AI shopping agents, across every selling surface. It works at the product level, upstream of ranking and paid placement. The two tracks do not change that definition. They double the number of places the work has to be done, which is the practical meaning of ACO Everywhere.
The first job is the Amazon listing. In our prompt study of Alexa for Shopping, the agent groups its picks into shelves by need, such as use case, product type and budget, and explains each pick with details it pulls from the listing: fit, materials, features. Leading a shelf is not the same as being the first pick, and our preliminary October audit found errors in roughly one in five Amazon product pages, most of them missing attributes and contradictory details. On this track the reader is Amazon's agent, the source is Amazon's listing fields, and the fix happens inside Seller Central or Vendor Central. We wrote about the holiday version of this job in "All Eyes on Amazon."
The second job is your own catalog, as the open web sees it. Muse's agents go to product pages and on-site search, build a shortlist and move the shopper toward checkout, and ChatGPT, Gemini and Google AI Mode do the same from the same data. On this track the reader is an answer engine, the source is your site, your feed and your structured data, and the fix happens in your PIM, your commerce platform and your product content. We wrote about this job in "Meta's Muse Is Already Shopping From Your Product Data."
Both jobs run through the same four questions an agent asks of a product: whether it knows what the product is, whether it can compare it against alternatives, whether it can reason about it against the shopper's constraints, and whether it has the full story. What differs is the source and the fix. The working principle is the same on both: treat the shopping agent like your best sales rep and arm it with everything you know about the product. A title tuned for Amazon's search ranking may not carry the attributes an answer engine needs, and a product page rich enough for Muse does nothing for Alexa if the Amazon listing is thin. Because Amazon says it will work with horizontal agents, the day is coming when one agent sees both versions of a product and compares them, and a contradiction between the two becomes a reason to leave the product out. On either track, inclusion is earned one SKU at a time, and a product the agent cannot interpret is silently excluded: left out of the answer rather than ranked lower.
Retailer agents read the retailer's listing. Answer engines read your catalog. Being ready for one does not make you ready for the other, and the two versions of a product need to agree.
Pick one product that matters this season and test it the way a shopper would, on both tracks, before you touch anything else. With under four weeks to Black Friday there is no time to re-enrich a whole catalog, so the useful move is narrow: one product, both tracks, the same questions.
On the Amazon track, open Alexa for Shopping and ask for the product the way a shopper would, by need rather than by name: the use case, the budget, the constraint that matters ("waterproof trail shoes for wide feet"). Note whether the product is included at all, which shelf it lands on, which details Alexa uses to explain the pick, and what it gets wrong or leaves out. Then put the listing next to that answer: the title, bullets, description and structured attributes in Seller Central or Vendor Central. Every detail the shopper asked for that the listing does not state, or states two different ways, is a gap.
On the answer engine track, ask the same questions in ChatGPT, Gemini and Google AI Mode, and in Muse if you have access. Check the same three things: whether the product appears as a product card, how the agent describes it, and which attributes are missing or contradicted. Then compare that against your own product page, feed and structured data, and against the Amazon listing, because the two versions of the product need to agree.
Fix attributes before copy. Completeness and consistency across both versions of the product come first, and richer descriptions and context come after. Then move to the next product on the list. The ACO Report scores a product on either track from its Amazon ASIN or its product page URL, but one product done by hand is enough to show where the gaps are and which track they sit on.
The Journal is right that the bots are coming. On either track, the first thing they reach is the product record, and the records that cannot explain themselves are the ones left out.
Request your free ACO Report for the product that matters most this season. Submit its Amazon ASIN and the report scores the listing Alexa for Shopping reads. Submit its product page URL and the report scores the page answer engines like ChatGPT read. Run the same product both ways and you have a readiness score for each track, the product details, page structure and trust signals that need attention, and a prioritized list of what to fix first for better visibility on both.