Until now, the rule was simple: if you sell online, you optimize your shop. With the Universal Commerce Protocol, a second sales surface is emerging that is not a shop at all – the conversation with an AI agent. Google established the standard together with Shopify, Target, Walmart, and other partners; it describes how agents query product data, compare offers, and initiate purchases. Time for a sober look at what this is – and what it is not.
The building blocks, sorted
| Building block | What it is | What it means for retailers |
|---|---|---|
| UCP | open standard for agent access to product data and commerce functions | the technical ticket into agent-driven purchases |
| AP2 (Agent Payments) | protocol for payments that agents initiate on behalf of users | checkout without a shop visit becomes possible |
| A2A / MCP | communication protocols between agents and systems | connecting your own systems to the agent world |
| Universal Cart | cross-channel shopping cart in Google Search | checkout moves into the search interface |
| Merchant Center expansion | dozens of new attributes: Q&A, accessories, replacement products | feed completeness becomes a ranking factor for agents |
What actually changes for retailers
The strategic shift happens before the click: a shopping agent makes a preselection from the data it understands – price, availability, attributes, compatibility, answered questions. Your own shop becomes the fulfillment layer for a decision that was made elsewhere. That is not a dystopia, but a relocation of the optimization work: from product page conversion to data completeness. Retailers whose feeds are already error-free and attribute-rich today have a head start that cannot be closed on short notice – data enrichment across tens of thousands of SKUs takes months, not sprints.
What to do now – and what not to
- Now: establish feed hygiene – error rate toward zero, mandatory attributes complete, monitor and populate the new Merchant Center fields.
- Now: answer product questions in structured form and model compatibilities as data – both are explicit new attribute fields.
- Now: complete the structured data on your product pages – agents read schema, not design.
- Not yet: building proprietary one-off solutions for every agent. The standard is young; those who bet on open protocols and clean data are prepared for whatever shape it takes.
The honest unknown: nobody knows how quickly agent purchases will gain volume – forecasts range from niche to channel revolution. The data work is not a gamble either way: complete, structured product data already improves search, feeds, and classic visibility today. Agent readiness is the bonus, not the bet.
You can find out whether your data is ready with the checklist in the article Why AI agents can't find your products – and with our data readiness audit on the PIM services page.
