Amazon is considering cataloging products across the web, including those it does not sell, in an internal project called World Product Graph. The initiative aims to let customers research and compare items without leaving the platform, and give artificial intelligence assistants a broader knowledge base for answering questions and making recommendations.
Reuters obtained an internal document of about six pages in PR/FAQ format, the type of memo Amazon uses to present initiatives, explain how they work, and assess their potential financial returns. The proposal falls under Selection & Catalog Systems, a division overseen by vice president Robert Tekiela. The document sets October 1, 2027, as the ideal announcement date, but does not confirm that the project will launch.
Amazon told Reuters that drafts may be outdated and offer an incomplete picture. The company added that the existence of a brainstorming document does not necessarily mean its current plans align with it. Therefore, World Product Graph remains an internal proposal: it could be changed, delayed, or never reach the market.
A catalog connecting versions and alternatives
The idea is to gather information about items even if they are not part of Amazon's inventory. The system would link products to their previous and subsequent versions, variants, features, and possible alternatives. That way, someone searching for a vacuum cleaner, for example, could compare models and find options sold only on a manufacturer's website, without having to piece together that information by consulting several stores and pages.
The project could also use purchase history and other data to tailor recommendations to customers' needs. The commercial goal is to keep some of the pre-purchase research process within Amazon: looking up specifications, differences between models, and reviews or alternatives that may currently lead users to Google, Reddit, or other sites.
The catalog would have two main ways of gathering data. Manufacturers could provide information and updates directly; Amazon, in turn, would use automated programs to crawl other retailers' pages and extract product data. The document notes that the system depends on figuring out how brands organize their catalogs on their own sites.
According to the memo reviewed by Reuters, Coca-Cola and Bayer have reportedly already joined the data-gathering effort, and Amazon has reportedly spoken with Procter & Gamble about a possible trial. None of those companies publicly confirmed the agreements described. It is also unclear how data would be obtained from sites whose operators do not participate directly, or under what circumstances their consent would be required.
Product searches start outside Amazon
The incentive to build a more extensive catalog is reflected in the search data cited by Reuters. A 2025 Morgan Stanley study estimated that, in the United States, 16% of online shoppers start researching a product on Amazon, compared with 57% who do so on Google. The distribution changes when people already know which item they want to buy.
| Starting point | Product research | Buying an item already chosen |
|---|---|---|
| Amazon | 16% | 41% |
| 57% | 23% | |
| Walmart | Not specified | 18% |
The difference points to two distinct moments: discovering what to buy and finding a specific item. Amazon is in a stronger position when users have already made their decision, but the internal document identifies the possibility that research starts outside the platform as a disadvantage. If World Product Graph gathers information about products Amazon does not sell, the company could compete for that initial stage and then direct shoppers to an offer available inside or outside its store.
Keeping searches on the platform also creates more opportunities to show recommendations and sponsored products. The project would not simply expand the store's product pages: it aims to turn the catalog into a source of information for customers and automated systems that compare options on their users' behalf.
Shop Direct already links to external retailers
Amazon has begun exploring this area with Shop Direct and Buy for Me. Shop Direct lets users find products on Amazon and visit the seller's website to buy them. In certain cases, Buy for Me uses an AI agent to complete the order on the external site once the customer confirms it; the retailer continues to manage inventory, shipping, returns, and customer service under its own terms.
On November 4, 2026, Amazon announced Shop Direct Merchant Workspace, a free portal for eligible merchants to connect their catalogs to the platform. The company says Shop Direct has helped tens of millions of customers discover hundreds of millions of products from external stores. The global catalog would extend that approach: it would not only make it easier to find items from other retailers, but also seek to structure product information from across the web.
These features are not available in all markets in the same way. Amazon announced Alexa for Shopping on Amazon.it on November 11, 2026, with features for finding information, comparing items, and checking price history. That announcement does not mean that all U.S. Shop Direct and Buy for Me features are available in Italy.
The challenge of Rufus and data quality
The memo also proposes improving Amazon's shopping assistant, now integrated into Alexa for Shopping and previously known as Rufus. According to the document, many of Rufus's answers were incorrect or incomplete, and running Amazon searches with the assistant was expensive. The proposal argues that a more comprehensive database could improve accuracy and allow less costly models to be used for some tasks.
The document put Rufus activity at about 83 million searches per week. Amazon says those figures are outdated and states that Alexa for Shopping currently achieves overall accuracy of around 97%. Reuters reports this figure as a company statement; it does not provide independent verification of how it was calculated.
Accuracy is no minor detail in product design: people consulting an assistant may not have enough knowledge to spot a wrong answer. A catalog connecting product pages, variants, and specifications could give the system more context, although its usefulness would depend on the information being well organized and up to date.
Shopping agents enter the fray
Amazon is also considering a scenario in which AI agents research and shop on behalf of their users. If those systems rely on a broad product index, controlling one of the most comprehensive catalogs could give Amazon a presence in searches that do not start in its own store. The competition already has a contentious dimension: in September 2026, the company blocked Muse, Meta's assistant, from accessing its store.
Reuters reports that Muse surpassed 10 million users in the United States and Canada since its launch in September 2026. JPMorgan estimated that it could become one of the most widely used consumer AI apps after ChatGPT. Amazon also initiated legal proceedings against Perplexity in November 2025 over the use of Comet, its AI-powered browser.
On August 4, 2026, a U.S. appeals court overturned the preliminary order Amazon had obtained in that litigation and sent the case back for further review. The decision did not definitively resolve the dispute. On the commercial front, CEO Andy Jassy said during the third-quarter earnings presentation that purchases made through external agents still accounted for a limited share; he also noted that Amazon's AI shopping tools had been used by more than 400 million people over the past year.
The tension lies in the fact that Amazon wants to gather information from other sellers' websites while trying to control how agents from rival companies access its store. Crawling public pages to gather data and using an account to complete purchases are different actions, subject to different technical, contractual, and security issues. The project has yet to clarify how it would address those limits.
A catalog to keep searches on Amazon
The historical precedent behind the idea is the Whole Earth Catalog, founded by Stewart Brand in 1968. It brought together books, tools, scientific equipment, and other resources, and explained where to get them. Its motto was “Access to Tools.” The parallel lies in gathering scattered information to help people make choices, although Amazon's goal is also commercial: to keep people researching before they buy on its platform and position itself for searches made by AI agents.
The final scope will depend on how Amazon turns the proposal into a product and what terms it sets for manufacturers and retailers. For now, the October 2027 date is only the one envisioned in an internal document, not a public launch commitment. Questions also remain about source selection, the order in which products would appear, and the space given to independent reviews.
If it comes to fruition, World Product Graph could make Amazon a place to look up even items sold elsewhere. The immediate question is whether the company will move from an internal proposal to announcing a catalog with global reach.
Sources: macitynet.it, The Verge
