Amazon's World Product Graph: every product, by October 2027
Amazon wants to be the answer to "what should I buy," not just the place where you buy it. An internal document reviewed by Reuters describes a catalogue of nearly every product sold anywhere on earth — including the ones Amazon doesn't stock, scraped from the retailers that do.
Amazon is building a project it calls the World Product Graph: a plan to hold data on nearly every product sold anywhere, partly by crawling rival retailers' websites. The document is written in the style of a press release and carries an aspirational public launch date of October 1, 2027, Reuters reported on Friday. It sits inside Amazon's Selection & Catalog Systems division and is organised under Robert Tekiela, a vice president in the unit. Amazon's statement to Reuters was carefully unenthusiastic: "Draft, outdated documents often paint an incomplete picture," a spokesperson said, and the existence of a brainstorming memo "doesn't necessarily reflect our current plans."
That sentence matters, because it is the difference between a product and a thought. The document is an internal PR/FAQ, the format Amazon uses to pitch a new service before anyone commits budget. People familiar with the matter told Reuters the project could be revised or shelved over cost or other concerns. Nothing has launched. What leaked is the argument for why it should.
The gap it is meant to close sits upstream of the purchase. Amazon wins when a shopper already knows what they want: 41% of product searches start on Amazon once the item is decided, against 23% on Google and 18% on Walmart, according to a 2025 Morgan Stanley report the document cites. It loses badly one step earlier, at research, where 57% of online shoppers begin on Google and only 16% start on Amazon. Tekiela's memo makes the case plainly — customers can always find products on Amazon, but "what's been missing is understanding: how do these products relate to each other, which one is right for me, and what else should I consider?" The answers, the document concedes, currently require leaving Amazon for brand sites, review sites and forums.
A graph of every product, its attributes and its relationships would let Amazon answer comparison questions inside its own walls, at the moment a purchase intent is still forming. Two streams would feed it: brands and retailers pushing product information continuously, and Amazon crawling other retailers' sites itself. Coca-Cola and Bayer have already agreed to contribute data according to the document, and Amazon held a "CEO-level P&G/Amazon summit" for a trial with Procter & Gamble. None of the three companies responded to Reuters' requests for comment.
The crawling half is where this gets legally interesting, and the memo reportedly does not say how it would work or whether Amazon would seek permission case by case. Amazon is suing Perplexity over an AI agent that shopped its site without permission last year, a case still ongoing, and it blocked Meta's Muse and xAI's Grok agents from Amazon.com on similar grounds. A company that treats automated access to its catalogue as trespass should expect Walmart, Target and every Shopify storefront to apply the same logic to theirs — and the long tail of global retail is a permanent maintenance bill, not a one-time crawl. The other risk is opt-in: once brands work out that their data improves Amazon's answer more than their own site's, some will decline to feed the graph.
Rufus is the second reason, and the more revealing one. Renamed "Alexa for Shopping" in May, the assistant handles roughly 83 million searches a week at considerable cost, and the document is blunt about quality: "many Rufus answers are factually wrong or incomplete, but customers lack expertise to detect these errors." Richer catalogue data, the thinking goes, would let Amazon run cheaper models and get more accurate answers out of them. Amazon disputes the framing, saying the usage figures are significantly outdated and that its own data puts Alexa for Shopping at about 97% accuracy worldwide. Andy Jassy has said more than 400 million people used the assistant in the past year. Two sets of numbers, one memo, no audit — but the telling detail is that a leaked internal document rates the company's own flagship AI product more harshly than the earnings call does.
The defensive motive is the one with money attached. Agents that search, compare and buy on someone's behalf could route shoppers to whichever retailer has the most complete product data, and Amazon's advertising business depends on shoppers staying inside its walls. Jassy downplayed the threat on the third-quarter call, saying "few customers are doing much shopping this way yet" and that advertising had barely been affected. The internal posture reads more cautious. Meta's Muse has passed 10 million users in the US and Canada since launching in September, and JPMorgan said in a research note that it "has the potential to become the most widely used consumer AI application since ChatGPT." Amazon blocked Muse from its retail site the same month; Elon Musk, whose Grok agent was also blocked, posted that maybe someone should "start a competitor to Amazon."
We have been down this road with Amazon and agents before — Amazon blocked AI agents for a year. Now it sells ads inside one traced the September turn from locking agents out to selling them placement. The World Product Graph is the layer underneath that move: rather than monetise the agent's attention, own the product facts the agent reasons over. Same strategy, one level deeper.
The counter-case is stronger than the leak implies. Amazon already runs the deepest catalogue in Western e-commerce, and structured product data is drifting toward commodity — Google Shopping has spent two decades normalising merchant feeds, and every serious agent platform now ingests them. Crawling is not a moat; a moat is a brand deciding that Amazon's answer is the canonical one. Brands have reasons to say yes (Amazon converts) and reasons to say no (Amazon competes with them). PR/FAQ documents are also cheap: Amazon writes them by the hundred and most die quietly in review. Read the leak as a statement of intent from the catalog organisation, not a roadmap the company has funded.
What gives the intent weight is that the plumbing already exists. Employees describe the project as an extension of two live services, Buy for Me and Shop Direct, which let customers buy items listed on other sites with Amazon acting as intermediary. Those products are the proof of concept; the graph is what happens if Amazon decides the middleman role should extend to knowing what everything is.
What to watch: whether Coca-Cola, Bayer or P&G say anything publicly about their data commitments; whether any large retailer objects to Amazon's crawlers, which would turn a product memo into a legal fight; and whether Alexa for Shopping's 97% accuracy claim ever gets an independent test. October 2027 is far enough out that the date signals intent rather than a schedule.
Which side of the graph would you rather be on — the retailer whose catalogue trains the assistant, or the shopper who never has to leave Amazon to compare? Write to us via our contact page.
