Amazon Plans ‘World Product Graph’ to Map Products It Doesn’t Sell
An internal Amazon plan seen by Reuters would index nearly every product sold online, part of an effort to keep shoppers and AI agents on Amazon.
Amazon has drawn up an internal plan to build a database of nearly every product sold anywhere on the internet, including items it does not carry itself, Reuters reported on Friday. The project, called the “World Product Graph,” is meant to keep shoppers from leaving Amazon to research purchases elsewhere. It would also help Amazon compete as AI shopping agents change how people shop online.
Reuters reporter Greg Bensinger based the report on an internal document and on interviews with people familiar with the matter. Amazon did not confirm the project. The company said draft documents often give an incomplete picture and that the brainstorming document does not necessarily reflect its current plans.
What the plan describes
The document is a PR/FAQ, the roughly six-page memo format Amazon uses internally to pitch new products as if they were already launching. Reuters said it gives an aspirational release date of Oct. 1, 2027. Not every PR/FAQ becomes a product, though, and the news agency said the plan could be changed or dropped over financial or other concerns.
According to the document, Amazon cannot be the place customers go for every shopping task if they must leave the site to learn about products. To close that gap, the project would gather product information in two ways:
- Brands would supply data about their products on an ongoing basis. The document says Coca-Cola and Bayer have signed up, and Amazon held what it called a “CEO-level” summit with Procter & Gamble about a trial. None of the three companies responded to Reuters’ requests for comment.
- Amazon would crawl other retailers’ websites on its own. Reuters could not determine how that crawling would work or whether Amazon would seek permission in every case.
XenoSpectrum, a technology news site that analyzed the report, said the system would also track the latest models and color variants and map how products relate to one another, such as which item replaces an older version. The effort sits within Amazon’s Selection & Catalog Systems unit and is led by Robert Tekiela, a vice president in that division, Reuters said. Tekiela did not respond to a request for comment. In remarks cited by Reuters, he described the challenge as less about helping customers find products and more about helping them see how products compare and which one suits them.
Why it matters
The plan points to a weak spot for Amazon in the early stages of shopping. Reuters cited a 2025 Morgan Stanley report that found only 16% of online shoppers begin researching products on Amazon, compared with 57% who start on Google. Amazon leads once shoppers know what they want, however, with 41% starting there versus 23% on Google and 18% on Walmart.
The rise of AI agents, software that can search and shop for a user, adds pressure. Amazon executives worry these tools could steer customers to rival stores with richer product data, Reuters reported. A complete product index could help ensure that agents working for shoppers look for items on Amazon rather than other retailers’ sites, which would also protect Amazon’s growing advertising business.
Amazon has resisted outside agents. It blocked Meta’s Muse agent from its retail site in September, the month Muse launched; Muse now has more than 10 million users in the US and Canada. Amazon also sued the AI search company Perplexity last year to block similar access, and that case is ongoing. After a post claimed xAI’s Grok agent had been blocked on Amazon, Elon Musk wrote on X last month that perhaps someone should build a competitor to Amazon, Reuters said.
Andy Jassy, Amazon’s chief executive, played down the threat on the company’s third-quarter earnings call. He said few customers shop through agents so far and that the advertising business had been minimally affected.
Problems with Amazon’s shopping assistant
The document also points to weaknesses in Amazon’s own AI shopping assistant, Rufus, which was renamed Alexa for Shopping in May. According to Reuters, it says many of the tool’s answers are wrong or incomplete, and users often lack the expertise to notice. It puts usage at about 83 million searches a week and calls those searches expensive to run. Better product data would let Amazon rely more on cheaper AI models, the document says.
Amazon disputed that description. The company said the usage figures are significantly outdated and that its data shows Alexa for Shopping is roughly 97% accurate worldwide. Reuters did not explain how that figure was measured, XenoSpectrum noted.
What comes next
Employees describe the project as an expansion of two existing US services, Reuters said. Shop Direct lists products from outside stores and sends shoppers to those sites, while Buy for Me lets Amazon’s AI complete purchases on another store’s site. XenoSpectrum, citing Amazon announcements, said Shop Direct covered more than 100 million products from over 400,000 outside stores as of March.
Many questions remain open, including whether merchants could correct or remove data collected by crawling and whether outside AI agents could use the index. Amazon is moving staff onto the project, Reuters reported, but it could not confirm whether the October 2027 target is realistic. Amazon shares fell just over 1% to $264.12 by midday Friday, during a broader market selloff that followed the resignation of Nvidia chief executive Jensen Huang earlier in the week.
