Short version: shoppers can now buy things inside an AI conversation, without ever visiting the merchant's website. OpenAI and Stripe built the Agentic Commerce Protocol (ACP), which powers ChatGPT's instant checkout. Google answered with the Universal Commerce Protocol (UCP), launched this January with major retailers and payment networks behind it. If the last two years were about being recommended by AI, this year the transaction itself started moving into the chat.
What is a commerce protocol, without the jargon?
A protocol is just an agreed format — a standard way for an AI agent to read your catalog, check price and availability, and hand a payment to your systems securely. Think of it like the card networks: you don't negotiate with every bank individually, you accept a standard and the ecosystem works. ACP and UCP are that standard for AI agents, and the assistants (ChatGPT, Gemini, Copilot, Perplexity) sit on top of them.
Do you need to "implement a protocol"?
Almost certainly not — and this is where the hype gets ahead of the reality. If you're on Shopify or a comparable platform, the platform speaks the protocols for you; selling through AI channels is increasingly a set of switches in your admin, not an integration project. The same is happening across the major commerce platforms. The plumbing is being handled.
The honest picture
The protocol layer is the platform's job. What the platform cannot do is make your product data good enough for an agent to confidently pick you. The toggle puts you in the room; your catalog decides whether you get chosen.
The part that's actually yours to fix
An AI agent completing a purchase takes on more responsibility than one making a suggestion — so it gets more conservative, not less. Before it spends a customer's money it needs to be sure: exact variant, true availability, accurate specs, honest pricing, clear shipping and returns. Catalogs with thin attributes, stale stock signals, or contradictory descriptions don't just rank lower in agentic checkout — they get skipped, because a wrong answer costs the assistant a customer's trust.
- Attributes an agent can match to the question ("waterproof", "fits a queen mattress", "nut-free")
- Variants and availability that are actually true when the agent checks
- Descriptions that state what it is, who it's for, and how it differs
- Structured data an agent can parse without guessing — across the whole catalog, not just bestsellers
What to do this quarter
Three things, in order. Turn on the AI channels your platform already offers — that part is nearly free. Test your own store the way we described in our catalog-visibility guide: ask ChatGPT, Perplexity, and Gemini your buyers' real questions and see if you exist. Then fix the data gaps, biggest products first. That last step is the whole game, and it's the one no toggle does for you.
Where to start
Our free AI Shopping Visibility Audit reads your catalog the way the agents do and hands you a prioritized gap list — so you fix what moves the needle instead of guessing.
