Amazon Plans a World Product Graph to Index Products Beyond Its Own Catalogue

Amazon is planning a data store called the World Product Graph, intended to hold information on every product sold anywhere on the internet, including products Amazon does not sell. The Verge reported the plan on 13 November 2026.

According to the same reporting, Amazon intends to make World Product Graph information available through its own site. The stated goal is to make shoppers less likely to leave Amazon to research products on Reddit or Google, where they currently go for independent reviews and comparisons.

The plan is tied to Amazon's AI shopping assistant. The assistant, now called Alexa for Shopping, was previously known as Rufus. An internal Amazon document on the project states "Many Rufus answers are factually wrong or incomplete," and describes running users' Amazon searches with AI as expensive. Reuters also reported on the project's internal framing.

The same internal document includes an October 1, 2027 release date for the World Product Graph. Reuters reported that it was unclear if or when the plan may be implemented. Amazon told Reuters that draft, outdated documents often paint an incomplete picture, and that the existence of a brainstorming document does not necessarily reflect current plans.

The broader context here is a familiar retrieval problem. An assistant is only as reliable as the structured data it can check its answers against, which is what engineers call grounding. Product listings vary from seller to seller, specifications are often missing or contradictory, and reviews are scattered across third-party sites. A graph is a database built from entities and the links between them, rather than rows in a table. A product graph resolves those messy listings into canonical product entities, with attributes and relationships attached, and that is one way to tackle the problem. The internal document's admission that many answers are wrong or incomplete suggests current grounding is not yet sufficient. The cost figure suggests that retrieval-heavy AI queries on the retail site carry a real per-query expense.

In my view, the more consequential question concerns scope. A graph covering products Amazon does not sell would make Amazon an index of the wider market, not only a marketplace. That changes what Amazon is to the shopper, and it changes what it is to the independent review sites and forums the plan explicitly seeks to displace. How Amazon would source or license data for products outside its catalogue will determine much of the plan's practical reach.

The timeline deserves the same scrutiny. An October 2027 date in an internal draft is a planning figure, not a commitment. Amazon has said such documents can be outdated, and Reuters reported that it is unclear if or when the project will proceed. Nothing announced so far is a launched product.

In my view, if the project does materialise, the benefit for shoppers is easy to see. A single, well-grounded answer on specifications, compatibility and price history could replace the dozen browser tabs many purchases now require. Two questions would remain open: whether that answer is accurate enough to trust, and whether shoppers want their research to come from the same company that sells the product.