AI Agents Browse Your Product Catalog

Executive Summary

Fact: Commerce (Nasdaq: CMRC), parent of Feedonomics, launched Agentic Catalog Exports (ACE) on April 27, connecting merchant product catalogs directly to OpenAI/ChatGPT, Google Gemini, Microsoft Copilot, PayPal, Stripe, Perplexity, and Amazon in a single feed.
Impact: Dell is already syndicating approximately 7,000 SKUs through ACE — laptops, servers, monitors, accessories — making its products discoverable by AI shopping agents without building seven separate integrations.
Surprise: The bottleneck isn’t access to these platforms; it’s data quality. AI agents surface products based on structured attribute completeness, not keyword density — a completely different optimization game than the one brands have been playing for two decades.

The conversation inside most marketing departments still centers on ChatGPT as a writing tool, or maybe a customer service bot. Meanwhile, the actual disruption is happening at a layer most brand managers haven’t looked at yet: product discovery. AI shopping agents are becoming a primary channel through which consumers find and evaluate products — and your product catalog either speaks their language, or it doesn’t.

On April 27, Commerce (the parent company of Feedonomics) made a move that should force that conversation. The company launched Agentic Catalog Exports, a service that lets enterprise brands syndicate their product data to seven major AI surfaces simultaneously — OpenAI/ChatGPT, Google Gemini, Microsoft Copilot, Amazon, PayPal, Stripe, and Perplexity — through a single, managed feed. Dell signed on as an early adopter, loading roughly 7,000 products.

One Feed, Seven Platforms: What ACE Actually Does

The problem ACE solves is boring-sounding but commercially significant. Every AI shopping platform has its own schema requirements, update cadences, and data formats. For a mid-sized brand with 2,000 SKUs, building a compliant feed for seven different endpoints isn’t a marketing project — it’s a six-month engineering effort, and it never truly ends because the platforms keep evolving their specs.

Feedonomics centralizes that. You maintain one structured product feed; ACE handles the transformation, enrichment, and syndication to each destination. Sharon Gee, SVP of product for AI at Commerce, put it plainly: “Agentic commerce is quickly shifting from experimentation to real-world application, and merchants need a reliable way to participate.”

What’s striking about this move is the choice of destinations. PayPal and Stripe aren’t search engines — they’re payment infrastructure that is quietly adding agentic shopping layers. Perplexity isn’t a traditional retail platform. The implication is that AI-driven product discovery won’t be confined to chatbot interfaces. It will surface everywhere a consumer is already transacting or researching.

Dell’s 7,000-SKU Test and the Data Quality Wall

Dell’s participation is instructive, not because of its scale but because of what it required. Paul Mansour, Dell’s global marketing director, said it without euphemism: “As AI agents become a more common starting point for product discovery, the quality and structure of product data matter more than ever.” Feedonomics didn’t just connect Dell’s existing catalog — it optimized and restructured it so the products are “accurately and completely represented within ChatGPT.”

That restructuring is the real work. AI agents don’t rank products on keyword relevance the way a search algorithm does. They surface products based on how completely and consistently the structured data answers a buyer’s implicit question. A server listing missing its rack-unit dimensions, power draw spec, or supported RAID configurations won’t get surfaced to the IT buyer asking a Copilot agent to configure an infrastructure stack. The product doesn’t rank lower — it simply doesn’t exist in that context.

Epinium data

Across the brand catalogs we audit on the Epinium platform, over 65% arrive with incomplete structured attributes on first review — missing GTIN mappings, absent dimension fields, or misaligned category taxonomies. In traditional search, thin product data caused a gradual ranking slide. In agentic discovery, it causes immediate invisibility.

This is the uncomfortable truth for brands that have relied on strong photography and good copywriting to win on digital shelves. Neither translates to agentic surfaces. What translates is machine-readable completeness: every attribute field populated, every variant mapped, every compatibility note structured into the schema the agent expects.

The Channel Strategy Implication Nobody Is Talking About

The instinct for most e-commerce teams will be to treat ACE as another feed management task — the way they treated Google Shopping feeds in 2012. That framing undersells the shift. Google Shopping put products in front of humans who then exercised judgment. Agentic shopping puts products in front of agents that exercise judgment on behalf of humans, often without the human reviewing a search results page at all.

Amazon projected last year that AI-driven discovery would account for 50% of its search activity by 2029. That estimate was made before ChatGPT’s shopping features launched. The timeline may be shorter. The brands that are structuring their catalog data now, building the habits and workflows to keep attribute completeness high across all SKUs, are building an infrastructure advantage that compounds. The brands treating this as a 2027 problem are not.

What we’re seeing at Epinium is a growing gap between brands that have invested in catalog infrastructure — full attribute coverage, real-time sync, structured variant data — and those running on manually updated spreadsheets or legacy PIM systems with 40% field completion. That gap was tolerable when humans were doing the browsing. It becomes structural when agents are.

For COOs evaluating where to put AI budget in the next 12 months, this is a concrete answer: catalog infrastructure isn’t a legacy back-office problem. It’s the foundation your AI discoverability is built on. Feedonomics just made the first moat-building move. The question isn’t whether to participate in agentic commerce channels — it’s whether your product data is ready when you do.

FAQ: Agentic Commerce and AI Product Discovery

No — that’s precisely what ACE eliminates. Commerce manages the platform relationships and schema requirements on the backend. As a merchant, you maintain one structured catalog feed through Feedonomics, which handles transformation and syndication to each destination. Each platform does have its own eligibility criteria and onboarding process, but the engineering overhead of maintaining individual integrations is Feedonomics’ problem, not yours.

Does catalog size matter? Is a 200-SKU brand worth connecting to these channels?

Catalog size matters less than catalog quality and category fit. A 200-SKU specialty brand with complete, structured attribute data in a high-consideration purchase category (electronics, home improvement, B2B equipment) will outperform a 50,000-SKU retailer with thin listings. AI agents are particularly effective for complex, research-intensive purchases where the buyer wants a synthesized recommendation — those are exactly the categories where complete structured data pays off immediately.

What if my competitor syndicates their catalog to AI agents before I do?

This is the right question to be asking. AI shopping agents, unlike Google’s algorithm, don’t crawl and discover on their own schedule — they surface what’s in the feeds they’ve been given. If a competitor’s catalog is in ChatGPT’s shopping layer and yours isn’t, every buyer who starts that research journey in ChatGPT is a buyer who never sees your products. The first-mover window on agentic catalog presence is probably 12-18 months before these integrations become standard table stakes.

Can I control which products appear in AI agent recommendations, or is it all-or-nothing?

You have significant control. Catalog exports can be segmented by product line, margin profile, inventory status, or any attribute in your feed. Many brands will want to prioritize high-margin categories or in-stock items initially. The ability to gate specific SKUs from specific channels (say, keeping B2B-only products out of consumer-facing AI surfaces) is part of why a managed feed infrastructure matters — you need that segmentation logic in one place, not replicated across seven manual integrations.

My products are on Amazon and Google Shopping already. Is that enough?

It depends on where your buyers are starting their research. Amazon and Google Shopping remain dominant, but ChatGPT’s shopping features and Perplexity’s product search are capturing meaningful share of high-intent research queries — particularly among buyers 35 and under who treat AI assistants as default starting points. The more important question is whether your product data is structured well enough for all these channels: brands that have invested in attribute completeness will port to new surfaces easily. Brands with thin listings will face the same rebuild challenge on every new platform that emerges.

Agentic commerce is no longer a forecast. It’s shipping infrastructure, with Fortune 500 companies building catalog feeds for it right now. The brands that frame this as a data quality project — and execute it that way — will hold the shelf space that matters in three years. The brands that wait will find the shelf already occupied.

Ready to audit your catalog for AI agent discoverability? Epinium’s catalog management platform helps brands structure, enrich, and syndicate product data across AI and traditional commerce channels. Discover how Epinium optimizes your catalog for the agentic era →

For broader context on how AI agents are reshaping enterprise workflows, see our earlier analysis on what agentic AI means for business.