Amazon Drafts 'World Product Graph' to Index External Products and Power AI Shopping
Leaked internal documents reveal Amazon's planned 'World Product Graph' to catalog web-wide shopping data, cut AI costs, and enhance its shopping assistant.
Amazon is reportedly outlining plans to construct an exhaustive index of online merchandise across the web, aiming to catalog items regardless of whether they are listed on its own marketplace. According to internal documentation obtained by Reuters, the initiative—referred to as the "World Product Graph"—is designed to aggregate product data across third-party e-commerce platforms and retain user traffic within Amazon's digital ecosystem.
Amazon's vision for a universal shopping index
The primary objective behind the World Product Graph is to prevent consumers from leaving Amazon to perform product research on platforms such as Google or Reddit. By building a comprehensive repository of web-wide product data, Amazon hopes to ensure that consumers searching for items sold elsewhere still land on its pages.
This index is also tailored to interface with automated AI agents. As autonomous shopping bots become more prevalent, Amazon aims to position its database as the primary index that these agents query. The company has already laid some groundwork for displaying external items through its "shop direct" feature, which launched last year to link out to products sold by outside retailers.
Addressing AI assistant inaccuracies and costs
Beyond consumer retention and AI agent discovery, the project addresses operational challenges facing Amazon’s internal conversational assistant, Alexa for Shopping (formerly known as Rufus). According to the Reuters report, an internal document notes that "Many Rufus answers are factually wrong or incomplete" and running user search queries through the AI system remains expensive.
By feeding a broader and more structured dataset into its AI architecture, Amazon intends to elevate the precision of its shopping assistant. The document indicates that expanding the underlying dataset could enable the system to deliver higher accuracy while relying on less expensive AI models, thereby reducing infrastructure costs associated with running heavy conversational models at scale.
Implementation challenges and Amazon's response
Deploying the World Product Graph requires significant data acquisition. The internal document explains that the project's success "depends on discovering how brands organize their products on their own websites," implying that Amazon would need to crawl third-party retail sites to structure the necessary catalog.
It remains uncertain if or when the project will be officially launched. While the internal file referenced a potential target date of October 1st, 2027, Reuters noted that the timeline and final execution remain unconfirmed.
In a statement provided to Reuters, an Amazon spokesperson downplayed the document's authority: "Draft, outdated documents often paint an incomplete picture and the existence of a brainstorming document such as this one doesn’t necessarily reflect our current plans."
What it means for developers
For developers building AI applications, e-commerce tools, or autonomous shopping agents, Amazon's initiative highlights the growing importance of structured data in improving AI reliability and managing inference costs. Incomplete or inaccurate domain data often forces developers to rely on larger, more expensive language models to maintain answer quality.
As engineering teams build out their own retrieval systems or experiment with different model architectures, managing API access and infrastructure costs remains a top priority. Developers looking to benchmark or integrate various large language models can try top AI models cheaply through one API at https://apixoai.online.
Amazon's strategy reflects a broader pattern in software architecture: enriching context data allows AI systems to achieve higher factual accuracy while transitioning production queries to lightweight, cost-effective models. If major e-commerce platforms move toward indexing external domain data to power AI agents, developers will also need to adjust how their systems organize, expose, and scrape product metadata to ensure compatibility with emerging AI shopping workflows.
Source: Amazon has a plan to index the world’s online shopping data — The Verge AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
