Shoppable in ChatGPT: whose catalog is your product in?

When a shopper asks an AI assistant to buy something, the answer comes out of somebody's catalog. That sentence is the entire strategic question in agentic commerce, and most retailers still haven't answered it. On July 22, Shoppable's Universal Checkout plugin went live in the ChatGPT Plugin Directory — a 500-million-product catalog with multi-retailer cart consolidation, sitting between AI shoppers and the brands they're buying from.

What actually happened

Shoppable's plugin is available to US ChatGPT users, who add it from the directory and invoke it by name with @Shoppable. It handles product discovery across retailers, consolidates items from multiple merchants into a single cart, and processes the transaction — so a shopper can assemble a multi-brand order in one conversation and check out once instead of bouncing through three storefronts.

For brands, the pitch is a single integration: one connection to the checkout backend, and your products become reachable inside the assistant. Shoppable says the underlying universal checkout technology is covered by four US patents built over fifteen years, and the company previously shipped a universal checkout Model Context Protocol server — the same open standard the rest of the agent ecosystem is converging on. The announcement pegs ChatGPT at roughly 900 million weekly users. Founder and CEO Heather Udo frames the thesis directly: "The cart should come to the consumer, wherever they already are."

Why it matters for your business

She's right about the direction, and that's exactly why the intermediary question matters. There are now three shapes of AI commerce, and they are not equivalent for you.

Direct rails — Square routing ChatGPT and Claude orders into your POS — keep the customer relationship and the order data on your side. Platform catalogs like Shopify's put your feed into a system you already operate. Aggregator catalogs get you reach fastest and cost you the most control: your product lives in someone else's index, merchandised by their logic, checked out through their flow. You may never learn the buyer's email.

None of those is automatically wrong. Reach is real, and a 500M-SKU catalog in front of a giant assistant is genuine distribution. But pick deliberately, and pick with a floor under you: your product data should be canonical in a system you own — clean titles, real attributes, accurate inventory, structured for machines — and then syndicated outward to every channel. Authoring your catalog inside an intermediary is how you end up unable to leave.

The tell is simple. If a channel disappeared next quarter, could you rebuild the feed from your own data in an afternoon? If the answer is no, that's not a sales channel. That's a dependency.

Sources: PYMNTS, MarTech Series.