Reuters reported on an internal Amazon document outlining the 'World Product Graph' plan: the company aims to collect structured data on virtually all products sold online, including items not in its catalog. The goal is to make this database the primary source for AI shopping agents, ensuring that even purchases made through third-party bots ultimately go through Amazon. There is currently no official confirmation of the plan.

image

What Happened

On November 13, 2026, Reuters published a report based on an internal Amazon document detailing the 'World Product Graph' project. The document cites a tentative launch date of October 1, 2027, and names VP Robert Tekiela as the project coordinator. Brands such as Coca-Cola and Bayer are already joining the program, and P&G held a meeting at the CEO level. At the time of the leak, the project had no public API, datasets, or methodology—only a draft was on the table. Amazon commented on the publication, stating that the document is an outdated brainstorming draft and does not necessarily reflect the company's current plans.

Context

The concept is driven by a shift in where shopping begins. According to Morgan Stanley data from 2025, 57% of users start general product searches on Google, while only 16% start on Amazon. AI assistants are playing an increasingly significant role in researching products on behalf of users; without addressing this shift, Amazon risks becoming merely the checkout point at the end of the funnel. The company has already established a defensive line for its territory: in September 2026, it blocked Meta's AI agent Muse and is suing Perplexity, while simultaneously developing its own shopping assistant. In this logic, the 'World Product Graph' is not about data quality for its own sake, but about controlling the entry point to agentic commerce: the owner of the graph decides who gets product information and on what terms.

Why This Matters for the Industry

For the industry, this is a competition for the data layer, not model architecture: the outcome will be determined by graph coverage, data freshness, and contracts with brands. If the plan is realized even partially, Amazon will become the de facto infrastructure layer for AI shopping agents and a platform for an entire class of products based on structured product data. However, this is also a classic red flag for platform dependency: building a product on top of a graph with undefined access terms is risky. For now, the only reasonable behavior for developers is not to build critical pipelines relying on such a source, but to monitor brand pilots and access term announcements. As a low-cost preparation, developers should normalize their own product feeds: schemas, identifiers, and attributes. In the coming months, the signal should become clearer: either official announcements, partner pilots, and initial API fragments will appear, or the project will fade into silence, confirming the draft version theory.

Why This Matters for Users

There are no direct changes for shoppers at the moment: the product does not exist. If the graph is launched, AI assistant responses about products will become more accurate and comprehensive, including Alexa for Shopping (formerly known as Rufus): the database promises to cover items not in Amazon's catalog, so recommendations will no longer be limited to its storefront. The downside is the dependency of the agents themselves: without access to the graph or on its terms, third-party bots lose utility, and the choice of independent shopping assistants may narrow to those with whom Amazon negotiates.

What Is Unknown / Limitations

Key caveat: the evidentiary basis for the publication is a single leaked excerpt from an internal document, the relevance of which the company itself has questioned. Nothing is known about technical feasibility: brand onboarding shows negotiation progress, but this is a marketing proxy metric, not evidence of a working data pipeline. Access terms for external agents are also undefined—whether it will be an open API, licensing, or a closed pilot. The stated launch date is tentative: the project may be delayed, scaled back, or remain on paper.

Sources

Author

Look at AI, editorial team