Amazon Plans a Database of Every Product Sold Online to Counter AI Shopping Agents

The project, called the World Product Graph, is described in a PR/FAQ, the mock press release and question-and-answer format Amazon uses to pitch new products internally. The document gives an aspirational release date of October 1, 2027. Reuters said it could not determine whether or when the plan would be carried out, and that it could be changed or abandoned.

What is at stake is control of the starting point of online shopping. If AI agents begin choosing where to buy on users' behalf, the platforms holding the most complete product data are best placed to capture that traffic. Amazon executives are worried that agents could steer customers towards sites with richer data, Reuters reported, citing people familiar with the matter.

According to the document, the graph would draw on two sources: product data contributed by brands, and Amazon's own crawling of other retailers' websites. It says the effort "depends on discovering how brands organize their products on their own websites." Reuters could not establish how the crawling would work or whether Amazon would seek permission.

Coca-Cola and Bayer have signed up to contribute data, and a "CEO-level P&G/Amazon summit" was held for a trial with Procter & Gamble, according to Reuters. None of the three companies responded to its requests for comment. The work sits in Amazon's Selection & Catalog Systems unit under vice president Robert Tekiela, who did not respond either, and staff are being reassigned to it. Employees described the project as an extension of Buy for Me and Shop Direct, existing features that list items sold on other websites.

The document is blunt about Amazon's own AI shopping assistant, Rufus, which was renamed Alexa for Shopping in May. "Many Rufus answers are factually wrong or incomplete, but customers lack expertise to detect these errors," it says. It puts usage at about 83 million searches a week and describes them as expensive to run. A larger store of product data would let Amazon answer more accurately and rely on cheaper models, the document suggests.

Amazon disputed that picture. "Draft, outdated documents often paint an incomplete picture," the company said in a statement to Reuters, adding that the existence of a brainstorming document "doesn't necessarily reflect our current plans." It called the usage figures "significantly outdated" and said, based on its own data, that Alexa for Shopping gives correct answers about 97% of the time worldwide. Chief executive Andy Jassy has said the assistant was used by more than 400 million people over the past year.

Amazon has taken a hard line with outside agents. It blocked Meta's Muse agent, which has gathered more than 10 million users in the US and Canada since its September launch, and it is suing Perplexity to stop similar access in a case that is still under way. After a post on X said xAI's Grok agent had also been blocked, Elon Musk replied: "Maybe we should start a competitor to Amazon."

The company starts from a mixed position. A 2025 Morgan Stanley report found that 16% of online shoppers research products on Amazon, against 57% on Google, although 41% of searches for purchases shoppers already intend to make begin on Amazon, compared with 23% on Google and 18% on Walmart.

For now, agent-led buying remains small. On Amazon's third-quarter earnings call, Jassy said "few customers are doing much shopping this way yet."