Universal Commerce Protocol: Get Your Feed Agent-Ready

Google's Universal Commerce Protocol (UCP) is about to add a new kind of shopper to your store: an AI agent buying on someone's behalf. If one tries to buy from you next month, it needs three things — a clean product feed, a verified Merchant Center account, and product schema that matches what a customer sees on the page. Get those right and an agent can add your item to a cart and check out. Get them wrong and it quietly picks the competitor whose data it could actually read.

UCP is the reason this matters now. It gives AI agents a standard way to discover products, compare them across shops, confirm price and stock, and complete a purchase for the shopper. The agent is a buyer with a checklist. Your job is to pass every check.

The good news: almost everything an agent needs is data you should already be maintaining. This is a cleanup job, not a rebuild. Below is what to fix, in the order that pays off fastest.

What the Universal Commerce Protocol actually changes

UCP turns your product data into the interface an autonomous agent reads before it decides whether to buy from you. Instead of a human scanning your page and forgiving a missing size chart or a slightly stale price, a machine parses structured fields and moves on the moment something doesn't add up.

The practical shift is that consistency beats cleverness. An agent cross-checks your feed against your live page and your structured data. If your feed says €39.00, your page says €42.00, and your schema says €39.99, the agent treats your listing as unreliable and downranks or skips it. A human would shrug; an agent disqualifies you.

It also raises the cost of the boring stuff: GTINs, availability, shipping and return terms. These used to be "nice to have" fields that improved your Shopping listings a little. Under an agent-led purchase, they are the difference between a completed order and a failed handoff. Treat every optional field as required until proven otherwise.

Fix your product feed first

Start with the feed. It is the single source most agents read before they ever touch your site, so a feed with correct identifiers, accurate prices and honest stock levels does more for agent-readiness than anything else on this list.

Work through these fields for every product:

• id — stable and unique. Never reuse an ID for a different product when a line is discontinued.
• gtin / mpn / brand — supply the real GTIN where one exists. Agents use it to match your item against other shops and reviews. No GTIN, no comparison, no sale.
• price and sale_price — match the live page to the cent, including currency and tax treatment for your EU market.
• availability — in_stock, out_of_stock or preorder, refreshed close to real time. Stale stock is the fastest way to fail a checkout.
• shipping — costs and delivery windows per country you serve.
• image_link — a clean, high-resolution main image with no promotional overlay text.

Set your feed to update at least daily, and hourly for price and stock if your platform allows it. If you run WooCommerce or PrestaShop, a scheduled feed export plugin handles this; on Shopify the native Google channel keeps it in sync. Whatever you use, confirm the update ran — a broken cron job that silently stops refreshing prices is a common and expensive failure.

Get Merchant Center clean and verified

Next, Merchant Center. This is where Google decides whether to trust your shop at all, which makes a clean account with zero disapprovals the entry ticket for agent-led buying. If products are disapproved there, no protocol will route a purchase to you.

Open the Diagnostics tab and clear every warning, not just the hard errors. The ones that quietly cost you sales:

• Missing or mismatched return and refund policies — agents want to know the terms before committing a purchase for someone.
• Unverified business information — name, address and contact details that match your live site.
• Microdata mismatch warnings where your on-page structured data disagrees with the feed.
• Shipping settings that don't cover the countries your feed claims to ship to.

Add your shipping and returns policies as structured settings, not just as a page of prose buried in your footer. Agents read the settings; they don't read your terms and conditions like a lawyer. Verify and claim your website URL so there is no ambiguity about which domain owns which listings.

Match your structured data to the page

Now line up your on-page Product schema with both your feed and what a shopper actually sees. That three-way agreement is exactly what an agent verifies before it trusts the listing. Contradictions are treated as errors, and errors get you skipped.

Use JSON-LD for Product and include a full Offer block. At minimum:

• Offer.price and priceCurrency — identical to the feed and the page.
• availability — using the schema.org values (https://schema.org/InStock etc.).
• priceValidUntil — so an agent knows how long to trust the price.
• shippingDetails and hasMerchantReturnPolicy — the same terms you set in Merchant Center.
• aggregateRating and review — only if they are genuinely on the page. Fake or orphaned review markup earns a manual penalty.

Test every template with Google's Rich Results Test and the Schema Markup Validator before you push it live. Then spot-check a real product URL, not just your homepage — the bugs always hide in the variant pages, the out-of-stock states and the sale-price logic. One extra step worth the five minutes: view the rendered page source in your browser and confirm the JSON-LD is present after JavaScript runs. Plenty of themes inject schema client-side, and if an agent reads the raw HTML before your scripts fire, it sees nothing at all.

A five-minute readiness check

You can gauge how agent-ready a single product is in about five minutes by comparing four things side by side. If all four agree, an agent can transact; if any disagree, fix it before moving on.

Run this on your five best-selling products today. If those pass, script the same comparison across your catalogue rather than eyeballing thousands of lines. The pattern of errors you find on the top five almost always repeats everywhere else.

Where hosting quietly makes or breaks this

An agent judges your shop partly on how fast and reliably it can read your data, so slow feed generation and flaky product pages can cost you sales even when the data itself is perfect. If a feed export times out mid-generation, Merchant Center simply keeps serving the last successful version — which may be hours stale on price and stock. And if a product page stalls or returns a 5xx under a burst of automated requests, an agent has no reason to retry when a competitor's page loaded fine. Neither is a certainty, but both are plausible ways good data still loses the sale.

This is where server headroom matters more than it did when only humans browsed. If your store is on shared hosting that struggles to build a large feed export, move the export to off-peak hours or upgrade the plan that runs your shop. At TPC Hosting our shop hosting is EU-based and GDPR-friendly, which keeps your customer and order data on this side of the border, and there are real engineers on support 24/7 if a feed job or a caching layer starts misbehaving at an awkward hour.

If you're moving a store to sort this out properly, migration is free and you have 30 days to back out, so you can test your feed and checkout flow on the new setup before you commit. Keep the pricing honest with yourself too: agent-readiness is maintenance, not a one-off project. Recheck the feed-page-schema match every time you change your theme, your prices or your product platform.

FAQ