Shopping agents · News from · Published · Stefan Hamann

Amazon is considering a global product index for AI search, Reuters reports

An internal concept called World Product Graph would bring together product information beyond Amazon. Whether or when it will launch remains open.

AmazonProduct dataDiscoveryTrust

A laptop product catalogue is connected to several independent shops and their products.
Illustration: Agentic Commerce

What Reuters reports

Amazon is examining how to gather information about products that are not sold on its own marketplace. On November 13, Reuters reported an internal concept called World Product Graph, citing a document it reviewed and people familiar with the effort.

According to the report, the proposed index would bring together data on a very large range of products. It describes potential contributions from brands and merchants as well as Amazon independently crawling other websites. Reuters could not determine how Amazon would crawl the sites or whether permission would be needed in every case.

Reuters identifies the document as an internal planning format used to evaluate an idea. Such documents do not necessarily lead to a launch, and the report says the project could be changed or abandoned. It does not establish that a World Product Graph is available to the public.

Why product knowledge matters to a shopping agent

An assistant needs to do more than repeat a product name. A customer may ask whether a coffee grinder fits an existing machine, whether a replacement part matches a particular revision or when an item will arrive. Answering requires relationships between products and dependable attributes. A broader collection could help with those questions. The report does not establish how current, complete or verifiable each piece of information in the proposed index would be.

Reuters says the internal document also discusses incomplete or inaccurate answers from Amazon’s assistant formerly known as Rufus. Amazon told Reuters that the usage figures in the document were substantially outdated and that its own data put Alexa for Shopping’s global accuracy at about 97%. The report does not explain the tasks or methods behind that figure, so the two statements cannot be compared directly.

A different proposition from the existing merchant feed

For Shop Direct, Amazon described a workspace in early November where eligible merchants can verify their website and connect a product catalogue. Customers in the US can discover those products through Amazon and Alexa for Shopping. A link can lead to the store; for eligible offers, Buy for Me can complete a purchase there after the customer confirms the details.

The World Product Graph described by Reuters is an internal concept, not an announced Shop Direct feature. Alongside possible data contributions, the report describes crawling other websites. It does not establish whether or how the two approaches would later work together. For a merchant, a deliberately maintained feed and information collected from public pages have different paths back to their source.

Who shapes the path to a decision?

A broad product index could help customers compare variants and offers from more sources. The interface also determines which information they see and how they reach an offer. Whether specialist advice, local collection or a repair service is visible depends on the available data and the way it is presented. The Reuters report does not establish a future ranking or how any individual merchant would be treated.

The question extends beyond Amazon. OpenAI’s Business Hub gives eligible businesses a place to maintain their own details and product feeds. The approaches differ, but they share a practical concern: a merchant needs to know which version of an offer an assistant sees and whether it gives the customer enough to make a good decision.

What merchants can check today

In Being found before anyone visits the shop, I use a suitcase whose dimensions including its wheels matter for a flight. The customer needs more than a search result. They need to know whether the product fits and how to reach a valid offer. A large index only helps with that job when the decisive details are correct and point to the right item.

A useful place to start is a small selection of products customers often ask about. Compare the precise variant, dimensions or compatibility, price, availability and delivery terms across the product page and existing feeds. Then try real customer questions in relevant search interfaces and record separately whether the product appears, whether the answer is correct and where the link goes.

Access deserves its own decision. Product search, model training and a specific purchase serve different purposes. The technical and contractual choices available to a merchant depend on the service involved. Reuters did not establish Amazon’s proposed crawling method or how consent would work in individual cases.

Source and further reading

Read the original announcement ↗

Read the related Agentic Commerce guide ↗

The Agentic Commerce handbook ↗

Save and share