Inside Amazon, the project has a name that sounds like a map: World Product Graph. The people working on it describe it in less poetic terms. The goal is a data entry for nearly every item for sale on the internet, including the items Amazon does not stock, does not ship and does not want.
Reuters reported on Nov 13, citing internal documents, that the effort is under way and that it goes well beyond the catalog Amazon already maintains. The company’s existing index covers what it sells. The new one would cover what exists, resolved into a single record per product, with attributes, images, prices and identifiers attached.
The distinction sounds academic until you consider who needs it. Shopping agents built by OpenAI and Meta are only as good as the product facts they can retrieve. If an agent is asked to find a specific jacket in a specific size at the lowest price, it needs to know that the jacket sold on five different sites is one object rather than five, and it needs structured attributes rather than five pages of marketing copy.
Amazon has spent twenty-five years building exactly that layer for its own store, and it is now proposing to build it for everyone else’s. The company would host the index and serve it through its own site, which means the reference point for product identity on the open web would sit behind a login box in Seattle.
According to a person familiar with the project, the internal framing is defensive as much as commercial. Agents are already sending traffic to merchants outside Amazon, and the company would rather supply the facts those agents use than watch a rival assemble them. The same person said no decision has been made on whether the graph will be offered to third parties through an interface, and that the question is being argued inside the company.
That question is the whole story. A product graph that stays inside Amazon is a better shopping experience for Amazon’s customers. A product graph that is published is infrastructure, and infrastructure is a position that is very hard to dislodge once it is adopted. The companies building shopping agents would have to use it, because the alternative is assembling the same data themselves at a cost that makes no sense for any single retailer.
A seller briefed on the plan said the prospect is unsettling for reasons that have nothing to do with competition. Amazon’s catalog rules are strict: standardized attributes, image requirements, title formats. Sellers who list only on their own sites have never had to meet them. If the graph becomes the index that agents query, the rules become the price of being found, and the price is paid by merchants who never signed up for Amazon’s marketplace.
There is a precedent, and it is uncomfortable for the company. Google built an index of the web and then had to defend, for fifteen years, the terms on which that index was monetized. Amazon would be building an index of commerce while operating the largest store in it, which is a combination that competition authorities in Brussels and Washington have already shown interest in.
Analysts said the timing reflects a genuine shift in how people find products rather than a defensive reflex. Search engines sent users to pages. Agents answer questions, and an answer requires a fact, not a link. Whoever holds the facts controls the answer, and the answer is where the purchase begins.
The company’s own marketplace gives it an advantage that is easy to underestimate. Amazon sees what people buy, what they return and what they search for and abandon. That behavioral data is what turns a static catalog into a ranked one, and ranking is the only part of the problem that is genuinely hard.
A competitor can crawl the web for product pages. It cannot crawl for the record of what happened after the click.
The project also fits an older pattern in Amazon’s history. The company has repeatedly built internal systems and then sold them: computing capacity became AWS, logistics became Fulfillment by Amazon, and advertising became a business larger than most of its retail categories. Each of those started as a capability built for the store and ended as a product sold to everyone the store competed with.
Whether the graph follows that path depends on a decision that has not been made, and on a legal environment that is less forgiving than it was when AWS launched. European regulators have spent years probing whether Amazon uses marketplace data to compete against the sellers who generate it. A graph that indexes those sellers’ products, ranks them and serves the ranking would be a new version of the same accusation.
For now the work continues quietly, one product record at a time. The company has not announced the project, has not described a launch date and has not said what it will charge, if anything. The documents reviewed by Reuters describe an effort measured in millions of records and years rather than quarters.
The interesting question is not whether Amazon can build it. It can, and it has most of the pieces already. The question is whether the rest of the web will accept a single company as the authority on what a product is, and what it costs, and whether it is in stock. That acceptance would be given quietly, by developers integrating an interface because it is easier than the alternative, and it would be very difficult to take back.
