Amazon's internal World Product Graph aims to collect data on all products online

An internal Amazon document reviewed by Reuters describes a plan called World Product Graph: to gather data on virtually every item sold online — including products Amazon doesn't stock.

File photo: Amazon's Day 1 tower in Seattle (file photo)
Amazons «Day 1»-tårn i Seattle, januar 2017. Filbilde. File photo: Adamajreynolds, CC BY-SA 4.0 · source This adaptation is licensed under CC BY-SA 4.0.

Amazon's internal World Product Graph aims to collect data on all products online

An internal Amazon document reviewed by Reuters describes a plan called World Product Graph: to gather data on virtually every item sold online — including products Amazon doesn't stock. The goal is to keep both customers and AI agents on Amazon.com, while solving quality and cost problems with the company's own AI shopping assistant. Amazon itself disputes that the document reflects current plans, and it is unclear whether the project will ever be realized.

What the document says

The exclusive was published by Reuters on November 13, 2026, and is based on an internal so-called PR/FAQ document — Amazon's standardized six-page pitch for new ideas, written as a fictional press release with an imagined customer as illustration. According to Reuters, the document belongs to Amazon's Selection & Catalog Systems division, organized under vice president Robert Tekiela. He did not respond to Reuters' inquiry.

The idea is large in its simplicity: to create the world's largest repository of product data, comprehensive enough to describe "virtually every item for sale on Earth, whether Amazon stocks them or not," as Reuters puts it, citing the document and people familiar with the matter. The document gives a desired release date of October 1, 2027, but labels it aspirational. Reuters could not determine whether or when the plan will be put into effect.

Amazon pushed back in a statement to Reuters: "Draft and outdated documents often paint an incomplete picture, and the existence of an ideation document like this doesn't necessarily mean it reflects our current plans."

Why Amazon wants this

The underlying motive is hinted at in a quote from the document: "Amazon cannot be the go-destination for every shopping mission if customers must leave the site to research product relationships on Google, brand websites, and review sites."

The numbers behind that concern come from a 2025 Morgan Stanley report cited by Reuters: 57 percent of online shoppers start their product search on Google, versus just 16 percent on Amazon. But as soon as shoppers have a specific item in mind, the order flips — Amazon is then the leading starting point with 41 percent of searches, versus 23 for Google and 18 for Walmart. In other words: Amazon dominates the moment of purchase but loses the research moment earlier in the funnel. That is precisely the gap a complete product graph is meant to close, so that customers never need to leave Amazon.com to compare, read about brands, or check reviews.

How it would work

The document describes two main mechanisms. The first is voluntary data contribution: brands and retailers continuously supply product data to Amazon. According to the document, Amazon has already signed Coca-Cola and Bayer as contributors, and the company reportedly held a "CEO-level P&G/Amazon summit" with Procter & Gamble about a pilot project. Notably: neither Coca-Cola, Bayer, nor P&G has confirmed their participation — the companies did not respond to Reuters' inquiries, so the information comes solely from the leaked document.

The second mechanism is more contentious: Amazon would, according to the document, independently crawl other retailers' sites. The document states that the project "depends on figuring out how brands organize their products on their own websites." How such crawling would take place, and whether it would require permission in all cases, is not clarified.

The AI connection: Rufus, Alexa and the costs

World Product Graph is not just a competitive project — it is also meant as a fix for the company's own AI shopping assistant, Rufus, now being carried forward as Alexa for Shopping. The document notes that "many answers from Rufus are factually wrong or incomplete," and that running users' searches through the AI is expensive. Usage is stated at roughly 83 million searches per week. More and better product data would, in theory, make the system more accurate and capable of running on cheaper models.

Amazon, however, rejects the premise. The company tells Reuters that the usage figures are "significantly outdated," and that Alexa for Shopping has an accuracy of approximately 97 percent globally. The two claims — the document's "factually wrong or incomplete" and Amazon's 97 percent — sit in unreconciled tension, and there is no public documentation that can settle the question.

The agent war: Muse, Perplexity and the wall around Amazon.com

A third dimension of the story is AI shopping agents — automated assistants that search, compare, and purchase on users' behalf, developed by other players. Amazon is already in conflict on this front: in September the company blocked Meta's Muse agent from its store, and litigation with Perplexity over the same kind of access is ongoing.

Here lies the strategic logic, according to people familiar with the plan, as quoted by Reuters: if Amazon owns the world's largest repository of product data, the agents will have to come to Amazon.com anyway to retrieve the information. That would not only give Amazon control over agents' access — it would also protect the company's rapidly growing advertising business, which is heavily tied to visibility in the platform's search results.

Why this might disappear

It is important to keep the skepticism in place. The PR/FAQ format is Amazon's standard way of pitching a new idea, and many such proposals never come to anything. The people Reuters spoke to said this project could be changed or dropped for financial or other reasons. The document also contains a fictional illustrative customer — a deliberate device of the format, but a reminder that the text is a thought experiment, not an implemented program.

Amazon also has a real interest in downplaying the leak, as any company would amid ongoing litigation and strained relations with Google, Meta, and Perplexity. The statement that the document "doesn't necessarily" reflect current plans deliberately leaves room on both sides.

Open questions

Several practical questions remain unanswered in the available reporting: How would items Amazon doesn't itself sell be ranked relative to the company's own assortment? Which countries would get the service first? What does the business model look like, and would crawling retailers' sites require consent in all cases? And a basic question about sources: the document itself has not been published, so all quotes and claims about Coca-Cola, Bayer, and P&G come from Reuters' review of it and cannot be independently verified from the leak alone.

What does still come through is the shape of the ambition: Amazon, which already dominates the purchase moment in online shopping, is considering — according to its own internal document — building a body of data that would make the research and agent-search moments too, and thus the shopping gateway of the AI future, Amazon's property.

AIMag.no
AIMag.no
The AIMag.no editorial team covers artificial intelligence, tools, research, and regulation.

Get the best of AI MAG in your inbox

News, analysis, and ideas at the intersection of AI and society.