Protocols · For Brands
Google's Universal Commerce Protocol (UCP), Explained for Brands
Google spent two decades teaching the world to shop through a search box. Now it is rebuilding that habit for a world where people ask Gemini and AI Mode what to buy. The Universal Commerce Protocol is how Google wants that purchase to finish inside the answer. Here is what it covers and what it asks of you.
What Google announced, in plain terms
Google introduced UCP as an open standard for agentic commerce, the layer that turns an AI conversation into an actual sale. The headline use case is direct buying inside AI Mode in Google Search and inside Gemini. You ask for a product, the assistant reasons about your request, and instead of bouncing you to ten retailer tabs, it can build a cart and let you buy in place.
Two details tell you Google is serious. First, UCP did not arrive alone. It is built to work alongside other industry standards, including a payments protocol for agents and the Model Context Protocol that AI tools use to call external systems. Second, it launched with broad backing across retail and payments, which matters because a commerce protocol only works if enough of the ecosystem actually speaks it. This is not a science project. It is Google trying to set the default for how agents buy things.
The piece that should change how you think: the Universal Cart
The most important idea in UCP is that the cart stops belonging to a single website. In the old world, a cart lived on your store, and a shopper had to be on your store to use it. In the agentic world, the cart can live inside the AI surface and still settle against your real systems. A shopper can assemble an order while chatting with Gemini, and that order flows back to you like any other sale.
If you want the one-sentence version of why this matters: the place where the buying decision happens is moving off your domain, and UCP is Google's answer for letting you capture the sale anyway. That is the same structural shift we describe in what agentic commerce is, expressed as a concrete Google product.
What UCP actually covers
You can reason about UCP through three capabilities. None of them are exotic. They are the minimum a machine needs to buy on your behalf.
| Capability | What it does | What it means for you |
|---|---|---|
| Cart | Lets the AI surface assemble and update an order in the conversation. | The shopper never has to land on your site to fill a basket. |
| Catalog | Lets the AI surface read your products and their attributes. | If your catalog data is thin, the assistant has nothing to reason with. |
| Identity linking | Connects the shopper's identity to a real, accountable purchase. | Returning customers and saved details work without you owning the chat surface. |
Notice what is doing the heavy lifting. Cart and identity are plumbing, and they are largely Google's problem to standardize. Catalog is your problem. If your product data is not rich enough for the assistant to compare and recommend, none of the transaction machinery helps, because you never make it into the answer in the first place.
UCP sits on top of Merchant Centre, it does not replace it
This is the single most useful thing for a brand to internalize. Google already understands your products through your Merchant Centre feed. That feed remains the foundation. UCP does not throw it away. It builds the transaction layer on top of the product understanding Google already has.
What changes is how much that feed is asked to carry. Feeds tuned for Shopping ads were built to win a click: a title, a price, an image, a few required fields. An assistant answering "which of these is best for a humid climate and sensitive skin" needs far more. Google has been expanding Merchant Centre with newer conversational attributes precisely for this. Five worth knowing:
- Q&A: the real questions shoppers ask, answered in your own data so the assistant can quote you instead of guessing.
- Document link: a pointer to deeper material, like a spec sheet or care guide, the assistant can draw on.
- Related product: explicit relationships between items, which helps the assistant build bundles and suggest alternatives.
- Popularity rank: a signal of what is actually selling, which feeds the assistant's sense of what to surface first.
- Variant option: clean structure for colour, size, and configuration, so "does it come in navy" gets a confident answer.
If those fields are blank, you are handing the assistant a thinner story than your competitor down the street. The work of filling them well is the same discipline we cover in product feed optimization for AI search and answer engine optimization for ecommerce.
UCP and ACP are two doors into the same room
Google's UCP and OpenAI's ACP are not competitors you choose between. They are parallel rails into different high-traffic surfaces, and the smart move is to be on both. The underlying work overlaps almost entirely, which is the good news.
| UCP (Google) | ACP (OpenAI) | |
|---|---|---|
| Main surface | Gemini and AI Mode in Search | ChatGPT |
| Feed basis | Merchant Centre plus conversational attributes | ACP product feed |
| Core capabilities | Cart, catalog, identity linking | Cart, feed, orders, authentication |
| Your real job | Rich, current catalog data | Rich, current catalog data |
Read the bottom row twice. Whichever protocol you start with, the work that decides whether you win is the quality and freshness of your product data. The protocols are how you connect. The data is whether you get recommended. We unpack the OpenAI side in the ACP explainer.
The part nobody warns you about: the spec keeps moving
UCP is new and evolving. Required fields, supported attributes, and surface behaviours change as Google learns what assistants actually need to recommend confidently. A field that is recommended today can become required next quarter. Keeping a feed perfectly aligned to a moving spec, across Google and every other AI surface, is not a project with an end date. It is an ongoing operation.
That is the honest reason brands lean on infrastructure here. Maintaining UCP readiness by hand means watching the spec, re-mapping your catalog, populating new attributes, and re-publishing, every time anything shifts. A managed layer absorbs that so your team is not stuck doing feed maintenance forever.
Show up on Gemini and AI Mode without the feed homework
Ziffi connects your catalog once, maps it to UCP and Merchant Centre's conversational attributes, and keeps it current as Google's spec changes. Nothing changes on your existing store. The integration is free, most merchants are live within hours, and Ziffi earns only on revenue it drives.
What to do first
Begin by measuring, not building. Find out whether Gemini and AI Mode recommend you today and where the buy link points, which is the whole subject of AI visibility and share of voice. Then audit your Merchant Centre feed for the conversational attributes above, and fill the gaps your competitors have already filled. Get those two things right and the UCP plumbing has something worth connecting to.
Common questions
Do I have to abandon Shopping ads for this?
No. Your Merchant Centre feed powers both, and AI Mode is an additional surface, not a replacement for the shopping experiences you already run. The richer you make the feed for AI, the better your other Google surfaces tend to perform too.
Is checkout actually happening inside Google?
That is the direction UCP and the Universal Cart are built for: the purchase completes in the AI experience and settles against your existing systems. Availability rolls out surface by surface and market by market, so the practical question is which experiences are live for your catalog, not whether the capability exists.
How is this different from how Perplexity recommends products?
Different sourcing, similar stakes. Perplexity leans heavily on cited web content alongside feeds, while Google leans on Merchant Centre. We compare the two demand-side surfaces in how to get recommended by Perplexity and Google AI Mode.