Amazon is secretly building a global product database to block AI shopping agents and protect its advertising revenue

Amazon is planning to expand its product database to encompass the entire internet, aiming to capture the entry point for shopping searches and defend against the impact of AI shopping agents on its traffic and advertising business.

On November 13th, according to an internal document obtained by Reuters, Amazon is advancing a project codenamed "World Product Graph," seeking to collect data on almost all products available for sale globally, regardless of whether those products are sold on its platform.

The goal of this project is to transform Amazon into a true “everything store” by possessing the most comprehensive product information, attracting users to complete the entire process from product research to purchase without leaving the platform.

The document is drafted in the form of a press release, setting October 1, 2027, as the target date for public launch. Amazon responded, stating that the document is a “draft, outdated document” and “does not necessarily reflect the company's current plans,” but also noted that the company continues to work on improving the comprehensiveness and accuracy of its product catalog.

The backdrop of this initiative is the rapid rise of AI shopping agents, which pose a potential threat to Amazon’s traffic and advertising business. Since its launch in September, Meta’s Muse shopping agent has accumulated over 10 million users in the US and Canada. Amazon soon blocked Muse’s access to its retail website and filed a lawsuit against AI company Perplexity last year, with the relevant case still under review.

Data Ambitions: Extending From Its Own Platform to All Online Products

According to the internal document reviewed by Reuters, the core concept of World Product Graph is to build a global product knowledge graph by understanding how brands organize product information on their official websites.

This involves two data acquisition approaches: first, inviting brands and retailers to proactively and continuously provide Amazon with product data; second, having Amazon independently crawl the websites of other retailers.

The document shows that Coca-Cola and Bayer have already signed up as data contributors, and Amazon has held a “CEO-level summit” with Procter & Gamble to promote pilot cooperation. The report has not yet confirmed how Amazon plans to crawl third-party websites or whether permission is required in all cases.

The project is part of Amazon’s Selection & Catalog Systems division, overseen by Vice President Robert Tekiela. Amazon is reallocating internal resources to drive project implementation, but the report stated it is unclear whether the October 2027 timeline is truly feasible.

Targeting Google: Competing for the Shopping Search Entry Point

One of the core motivations behind Amazon’s push of this plan is to close the significant gap with Google during the product research stage of the shopping process.

According to a 2025 report by Morgan Stanley, among consumers who research products before shopping online, 57% choose Google, while only 16% choose Amazon.

The internal document states bluntly: “If customers have to leave Amazon to research associated product information on Google, brand websites, or review sites, Amazon cannot be the top destination for every shopping scenario.”

However, when consumers have already decided to make a purchase, Amazon holds a clear advantage: 41% of searches start on Amazon, compared to 23% on Google and 18% on Walmart.

The goal of the World Product Graph is to extend this advantage from the purchasing decision phase forward to the product research stage, allowing consumers to gather information and compare products on Amazon even before they form a buying intention, thus expanding its influence in the shopping decision chain.

Fixing Rufus’ Shortcomings: AI Search Accuracy in Question

World Product Graph is also seen as a solution to the accuracy issues of Amazon’s own AI search engine Rufus (renamed “Alexa for Shopping” in May this year).

The internal document admits: “Many of Rufus’s answers are factually incorrect or incomplete, but users lack the expertise to spot these errors.”

The document also reveals that Rufus processes around 83 million searches weekly at a “high” operational cost. Amazon believes that richer product data will enable the use of more cost-effective models, thereby improving operational efficiency.

In response, Amazon stated that the usage data cited in the document is “severely outdated,” and its internal data shows Alexa for Shopping has about a 97% accuracy rate globally.

Amazon CEO Andy Jassy also downplayed the influence of AI agents on the third-quarter earnings call, noting “users who shop in this manner are still very few,” and that the impact on the advertising business is limited.

Defending Against AI Agents: Data as the Moat

Although Jassy’s public statements have been relatively low-key, according to insiders cited by the report, Amazon’s senior management remains highly alert to the possibility that AI shopping agents may direct user traffic to competitor sites with more complete product data.

Meta’s Muse agent was described in a JPMorgan research report as “the consumer-grade AI application with the most potential to become widely used since ChatGPT.” Last month, Elon Musk responded on X to reports that Amazon had blocked access to its xAI Grok agent by saying, “Maybe we should create an Amazon competitor.”

According to insiders, Amazon seeks to ensure that by building the world’s largest product database, AI agents tasked with shopping on behalf of consumers will ultimately need to return to Amazon to access products, rather than directing traffic to other retailers’ websites.

This strategy is also closely bound to the future of Amazon’s advertising business. By mastering more comprehensive product information, Amazon hopes to remain present throughout the product discovery, comparison, and purchasing process for consumers, preventing AI agents from bypassing its platform and weakening its control over shopping traffic and advertising revenue.

Employees describe World Product Graph as an initial attempt to make Amazon a “one-stop platform that sells everything,” and also an extension of the existing “Buy for Me” and “Shop Direct” services. These two services already allow users to purchase products from other websites through Amazon, with Amazon serving as the intermediary.