Analysis

Amazon's “World Product Graph”: an internal plan to catalog products it does not sell, including at rivals, with AI shopping agents in its sights

An internal Amazon document seen by Reuters plans a catalog of almost every product for sale, Amazon's or not. Amazon calls it an outdated brainstorm.

By ·

Amazon is considering building a database of "virtually every item for sale on Earth, whether Amazon stocks it or not", according to an internal document reviewed by Reuters and people familiar with the matter. The project is called World Product Graph. Reuters published the story on November 13 (CNA, carrying Reuters). Amazon replied that "draft, outdated documents often paint an incomplete picture" and that this brainstorming document "doesn't necessarily reflect our current plans".

Key points

What the document says

The main argument is that shoppers leave Amazon to compare products. "Amazon cannot be the go-destination for every shopping mission if customers must leave to research product relationships on Google, brand websites, and review sites," the document says, as quoted by Reuters. The project "depends on discovering how brands organize their products on their own websites". It sits in the Selection & Catalog Systems unit, under vice president Robert Tekiela.

The goal is to know how products relate to each other: which model or colorway is the latest, which one suits a given shopper. "Customers have always been able to find products on Amazon," Tekiela writes in the document. "What's been missing is understanding."

Rufus, an assistant the document calls unreliable

The document also targets Amazon's own assistant, Rufus, renamed "Alexa for Shopping" in May. "Many Rufus answers are factually wrong or incomplete, but customers lack expertise to detect these errors," it states. It counts about 83 million searches a week and describes running them as "expensive" (Reuters). More complete data would let Amazon use less expensive models.

Amazon disputes these figures. In its statement to Reuters, it says "the usage figures are significantly outdated" and that its own data puts Alexa for Shopping at roughly 97% accuracy globally. The document is not dated in the coverage, so we cannot say when its error assessment applied.

Why AI agents change the stakes

Why AI agents change the stakes
Data pointFigureSource
Shoppers who start product research on Amazon / Google16% / 57%Morgan Stanley, 2025, via Reuters
Searches for a known product starting on Amazon / Google / Walmart41% / 23% / 18%Morgan Stanley, 2025, via Reuters
Rufus searches per week (internal document)about 83 millionReuters; Amazon says the figures are outdated
Alexa for Shopping users over the past yearmore than 400 millionAndy Jassy, via Reuters
Meta's Muse agent users (US and Canada, since September)more than 10 millionReuters

Table source: CNA, carrying Reuters, November 13, 2026.

Amazon blocked Meta's Muse agent from its retail site in September, and last year sued Perplexity to prevent similar access, a case that is still ongoing, according to Reuters. In October, Elon Musk replied to a post saying xAI's Grok agent was blocked on Amazon: "Maybe we should start a competitor to Amazon." On the third-quarter earnings call, Andy Jassy said "few customers are doing much shopping this way yet".

Walmart has taken the opposite approach, opening an MCP server to approved agents (our October 31 article on the Walmart Agent Gateway). Read in that light, the World Product Graph looks like a third way: rather than open the door or block it, become the reference catalog. According to Reuters' sources, the most complete product data would compel agents shopping for users to look for products on Amazon.com rather than other retailers' sites, which would also protect its advertising business.

Our reading: the battle moves to data

For AI agents, catalog quality matters more than storefront design. An agent compares specifications, versions and prices; it goes wherever the data is most complete and most reliable. Amazon is acknowledging, in an internal document, that its own assistant falls short on this point, and is trying to turn the gap into an advantage. The project extends existing features, Buy for Me and Shop Direct, which already list products sold on other sites, according to Reuters and The Verge.

Two questions remain open. The first is legal: crawling competitors' sites at scale raises questions of permission that Reuters could not answer. The second is about neutrality: a catalog of "everything" run by a retailer that sells ads is not a neutral reference for an agent supposed to defend the buyer's interest.

What is not established

Method and limits

Sources checked on November 14, 2026. We did not read the internal document: all the quotes come from Reuters, read in the version published by CNA, and The Verge's write-up. The link between the project and the defense against AI agents comes from people cited by Reuters, not from quoted passages of the document. This is not investment advice.

Sources

Spotted an error in this article? Report it · Our corrections policy

OrigoTech logo

New: the app iPhone & iPad

OrigoTech is now an iPhone and iPad app

Read the articles natively, with “New” markers on unread ones, local search and offline reading of what you have already loaded. Free, no account, iOS 17 or later; ads appear between articles.

Rekha icon: a house drawn in a single stroke

From the same author Self-promotion

An Eulerian path, drawn with your finger

Rekha, the iPhone game by OrigoTech's author: connect every dot in a single stroke without tracing the same link twice, across 500 drawings that are all solvable.

The OrigoTech newsletter

A sourced weekly digest of AI tool news.

Read next