Amazon Secretly Building Global Product Database to Block AI Shopping Agents
nashnova research
Amazon is secretly developing a project codenamed "World Product Graph" — a database of virtually every product sold online, on or off its platform — targeting an October 2027 launch, a move aimed at locking down product-data control before AI shopping agents siphon away its traffic.
What exactly is the "World Product Graph"?
Amazon wants to catalog every product sold online, not just items listed on its own marketplace — if a product exists on any retailer's site, Amazon wants it indexed.
Two data pipelines feed the project: brands voluntarily contributing data (Coca-Cola and Bayer have signed on; Procter & Gamble held a "CEO-level summit" to pilot) and Amazon scraping third-party retail sites on its own.
This means → Amazon is trying to evolve from "a place to buy" into "the central hub for all product information online" — you don't have to sell on Amazon, but your product data lives there.
The project sits inside Amazon's Selection & Catalog Systems division, led by VP Robert Tekiela, with internal staff being reassigned to push it forward.
Why the urgency right now?
The immediate trigger: the rise of AI shopping agents. Meta's Muse shopping agent has accumulated over 10 million users in the US and Canada since launching in September.
Amazon's response has been aggressive: it blocked Muse from accessing its retail site, sued AI company Perplexity last year (case still pending), and blocked Elon Musk's Grok agent. Musk responded on X: "Maybe we should create an Amazon competitor."
In plain terms = AI agents let users compare and shop across platforms without opening Amazon first. Whoever controls the most complete product data feeds the best agent — and Amazon does not want that to be someone else.
JPMorgan called Muse "the most likely consumer AI application to achieve ChatGPT-scale adoption" — that verdict may be the real pressure behind Amazon's acceleration.
Where is Amazon losing the shopping journey?
Morgan Stanley's 2025 research shows: among consumers who research products before buying online, 57% turn to Google; only 16% turn to Amazon.
But when consumers already know what to buy, Amazon leads: 41% of searches start on Amazon, versus 23% on Google and 18% on Walmart.
This means → Amazon wins at "buy" but loses badly at "browse and compare." The World Product Graph aims to push Amazon's advantage backward from the purchase stage into the research stage.
The internal document states plainly: "If customers must leave Amazon for Google, brand sites, and review sites to research products, Amazon cannot be the default destination for every shopping occasion."
Why does Amazon's own AI search need this database too?
Amazon's in-house AI search engine Rufus — renamed "Alexa for Shopping" in May — has an accuracy problem. The internal document concedes: "Many Rufus answers are factually wrong or incomplete, but users lack the expertise to spot the errors."
Rufus handles roughly 83 million searches per week, and its operating costs are described as "expensive." Amazon believes richer product data would let it run cheaper models while delivering equal or better results.
In plain terms = the more complete the data, the less the model needs to guess — that improves accuracy and cuts costs by swapping in a smaller model.
Amazon countered that the usage figures cited are "significantly outdated" and that internal data shows Alexa for Shopping runs at about 97% accuracy globally.
Can this plan actually ship?
Amazon officially called the document a "draft and outdated" file that "does not necessarily reflect current plans" — while also stressing its ongoing commitment to catalog completeness and accuracy.
Reuters reports that whether the October 2027 target date is realistic remains unconfirmed.
This reflects a deeper tension: CEO Andy Jassy downplayed AI agents on the Q3 earnings call, saying "the number of people shopping this way is still small" — yet according to Reuters sources, senior executives are on high alert about agents redirecting traffic to competitors with more complete product data.
Put simply = the public message says "no rush"; the internal action says the opposite. That contradiction itself signals the threat is real.
市场有风险，内容仅供研究参考，不构成投资建议。