Cosmo's Backend Data Model: The 18 Critical Fields Amazon Rufus Actually Reads
Thursday, March 26, 2026
Amazon Built a Wall to Keep AI Agents Out. Your Listings Are on the Wrong Side of It.
In March 2026, a federal judge issued a preliminary injunction blocking Perplexity's Comet browser from accessing Amazon's platform. Amazon had sued the AI startup months earlier, and a court agreed: even if a shopper explicitly authorized Comet to shop on their behalf, that permission didn't extend to Amazon. The platform's authorization was a separate matter entirely.
That legal distinction is doing a lot of work right now. And most Amazon sellers have not thought through what it means for their business.
The case is a legal footnote in a much larger structural shift: the discovery layer above Amazon is being rebuilt by Google, OpenAI, Meta, and others. Amazon has responded by locking its doors. Forty-seven external AI agents have been blocked from crawling Amazon.com, including ChatGPT, Gemini, and Meta AI. Amazon updated its Business Solutions Agreement in early March 2026 to require all AI agents to formally identify themselves before accessing its services. Perplexity alleged in its defense that Amazon's real motivation was protecting its $56 billion advertising business, not customer security. The court found enough merit in Amazon's position to grant the injunction anyway.
None of that corporate warfare is the point. The point is what it means for your product's discoverability over the next 24 months.
How Shoppers Are Starting to Search
The number of shopping-related queries processed by AI agents is not small. OpenAI's Economic Research team published a working paper estimating that ChatGPT handles approximately 50 million shopping-related queries per day. McKinsey projects agentic commerce could account for $1 trillion in U.S. retail revenue by 2030. Morgan Stanley analysts expect nearly half of American shoppers will use AI agents by then.
These are not niche users. Research from Walmart's EVP of design and product found that ChatGPT's user base skews toward higher-income, technically sophisticated shoppers who are not traditional Walmart customers. The people most likely to try AI-powered shopping are, by demographic profile, the exact buyers premium Amazon brands want to reach.
When one of those shoppers opens ChatGPT and asks "what's the best magnesium supplement for sleep?" or "find me a chef's knife under $150 that won't need constant sharpening," the agent searches the open web and available commerce platforms. Amazon's catalog is not available to that agent. The brands that appear are the ones whose product data exists somewhere the agent can reach.
Amazon-only sellers are structurally invisible to this search.
The Walled Garden Is a Strategic Bet That May Not Pay Out for Sellers
Amazon's rationale for blocking external agents is coherent from a platform perspective. The company's advertising business depends on shoppers arriving at Amazon.com directly, browsing sponsored placements, and converting there. When an AI agent intermediates that process, sponsored listings disappear from the interaction. Amazon loses the ad impression. For a business generating $56 billion in annual advertising revenue, that is an existential concern.
The strategy makes sense for Amazon. It is not obviously good for the brands selling on Amazon.
Consider the comparison with Walmart. While Amazon has hardened its defenses, Walmart has moved in the opposite direction. Walmart embedded its AI assistant Sparky directly inside ChatGPT, enabling multi-item cart support, account integration, and Walmart+ loyalty syncing within the chatbot interface. Walmart's reasoning: ChatGPT is delivering new customers at twice the rate of traditional search. The demographic shift justifies the integration.
Amazon's subsidiaries tell the same story from inside the company. Zappos, Shopbop, and Woot do not block AI agents in their robots.txt files. The products discoverable there can be found and recommended by external AI tools. Amazon's main storefront cannot.
This creates a structural tension for sellers who exclusively distribute through Amazon: their products cannot be found by the fastest-growing segment of AI-powered product discovery, while competitors with Shopify storefronts or Walmart.com presence are indexed and available.
Google Is Building the Rails Faster Than Anyone Expected
While the Amazon-Perplexity lawsuit generated most of the press, the more consequential development for sellers happened quietly in late January 2026 when Google announced Universal Commerce Protocol (UCP). Think of UCP as standardized infrastructure for agentic shopping: a set of APIs that allow AI shopping assistants to check live inventory, compare prices across retailers, add multiple items to a cart, and complete purchases on behalf of the user.
The pace of Google's execution has been unusual. UCP was announced in January, went live in February, and received four major upgrades in March: multi-item cart support (the exact feature that killed OpenAI's Instant Checkout experiment), a live catalog endpoint with real-time inventory and pricing, loyalty program integration so a shopper's Ulta Rewards or Walmart+ membership travels with them into AI-powered shopping, and a self-service Merchant Center onboarding so any retailer can join without a partnership deal.
Sellers with Shopify stores can now connect to UCP through Google's Merchant Center and have their products surfaced inside Gemini-powered shopping experiences. Amazon sellers with no external storefront have no equivalent path.
Product Data Quality Is the New Moat
Here is the mechanism sellers need to understand. AI shopping agents do not operate like human browsers. They do not scroll product pages, evaluate images, or read between the lines of marketing copy. They query structured data: catalog attributes, normalized specifications, review sentiment signals, and metadata fields. The agents that run on UCP, ACP (Shopify's Agent Commerce Protocol), and similar standards pull from these structured sources to build recommendations.
This is the same underlying architecture that drives Rufus inside Amazon. As documented in Amazon's own technical publications, Rufus runs a two-stage retrieval process. In the first stage, it queries structured catalog fields to narrow products from millions to hundreds. In the second stage, it applies semantic matching on unstructured content like reviews and descriptions. A product with incomplete backend attributes gets filtered out in stage one before Rufus ever evaluates the copy.
The same logic applies outside Amazon. A seller whose product description says "great for outdoor use" but has no structured activity or use-case attributes will not surface in an AI agent query for "portable water filter for backcountry camping." The agent cannot infer; it queries.
The brands winning in agentic commerce are the ones treating product data as infrastructure, not marketing. Complete specifications, standardized measurements, accurate category mapping, and detailed review content are the inputs that AI systems use to determine relevance. This is not a future problem. McKinsey's February 2026 research found that electronics (38%), beauty and personal care (32%), and supplements (33%) are already leading categories for AI-assisted shopping decisions. These are mainstream Amazon categories with established seller bases who are, right now, invisibly losing discovery to better-structured competitors on open platforms.
The Practical Implication for Established Sellers
The structural shift does not require abandoning Amazon. It requires rethinking the assumption that Amazon is the only discovery surface that matters.
The actionable moves vary by situation. For sellers with Shopify storefronts, connecting to Google's UCP through Merchant Center is now a self-service action. For sellers on Walmart.com, Walmart's Sparky integration with ChatGPT provides passive AI-driven discovery for any product in that catalog. For Amazon-only sellers, the primary near-term lever is ensuring product data quality inside Amazon is complete enough to perform well inside Rufus, which is the one AI shopping surface they can still influence and measure.
That last point matters more than most sellers recognize. Rufus already accounts for substantial incremental revenue inside Amazon. Amazon confirmed in Q4 2025 earnings that Rufus was tracking toward nearly $12 billion in annualized incremental sales. The sellers capturing that are the ones whose structured catalog data is complete and whose review content is detailed. The sellers missing it have the same fundamental problem as sellers invisible to ChatGPT: their product data is not good enough for AI systems to confidently recommend them.
The underlying challenge is consistent whether the AI is inside Amazon or outside it. Your product data is your new storefront. Amazon's walled garden is a corporate strategy, not a seller benefit. The brands that treat data quality as infrastructure now will have a compounding advantage as the discovery layer continues to shift.
If you want to understand how your listings are performing inside Rufus specifically, Atomic AMZ offers a free Rufus audit atatomicamz.com. It is worth knowing before the next wave of agentic commerce makes the gap harder to close.
Further Reading
Frequently Asked Questions
What is agentic commerce?
Agentic commerce refers to AI-powered shopping experiences where an assistant browses, compares, and purchases products on a shopper's behalf. Rather than typing keywords into a search bar, the shopper tells an AI what they need and the agent handles discovery and checkout.
Why is Amazon blocking AI shopping agents?
Amazon has blocked 47 external AI crawlers including ChatGPT, Google, Meta, and Perplexity from accessing its platform. The primary motivation is protecting its $56 billion advertising business, which depends on shoppers browsing Amazon directly rather than being directed by a third-party AI.
How does Amazon's walled garden strategy affect sellers?
When AI agents from ChatGPT, Gemini, or Meta AI cannot access Amazon's product catalog, Amazon-only sellers become invisible to an increasingly large segment of shoppers who start product searches in those AI tools. Sellers with multi-channel presence on Shopify or Walmart.com retain surface area in AI-powered discovery.
What is Google's Universal Commerce Protocol (UCP)?
UCP is Google's infrastructure for agentic commerce, allowing AI shopping assistants to check inventory, compare prices, add to cart, and complete purchases across participating retailers in real time. Google launched UCP in January 2026 and released four major upgrades in March 2026, including multi-item cart support and loyalty program integration.
What should Amazon sellers do to prepare for agentic commerce?
Sellers should treat product data quality as a first-priority task. AI agents pull from structured catalog attributes, reviews, and product descriptions. Sparse or incomplete listings will not appear in AI-generated recommendations regardless of PPC spend. Ensuring completeness in category-specific backend fields, standardized measurements, and detailed review content positions products for discovery across multiple AI platforms.
Find out if your Brand is invisible to Amazons Rufus AI discovery tool and how to close the Gaps
