Shopify UCP: what Shopify already does, and what's still yours

AI shopping agents now have a standard way to talk to online stores. The Universal Commerce Protocol (UCP), co-developed by Shopify and Google and announced in January 2026, lets an agent find a store's products, build a cart and prepare a checkout without scraping pages. Google's Universal Cart, which launched in the US on 15 October, runs on it.

The obvious question for a merchant is what they need to do about it. For a Shopify store, the honest answer is: much less than the "agent-ready" checklists suggest. Shopify has already built and switched on the machinery. What it can't supply is your store's own rules and data, and that is where most stores have gaps.

To see where those gaps actually are, we probed 200 Shopify stores on 22 October 2026. The numbers below come from that probe unless a source is linked.

How we tested

We took the 1,000 Shopify stores with the highest estimated traffic in a database of Shopify stores we maintain, left out headless storefronts, and picked 200 at random. For each store we made only read-only requests:

• the UCP profile at /.well-known/ucp
• the /agents.md file
• /robots.txt
• four buyer questions put to the store's public policy tool, the one AI agents call to ask about returns, shipping and cancellations
• one product page, checked for structured data and Shopify's WebMCP script

We did not create carts or checkouts. A Search Engine Journal study of 122 stores did, and found the checkout itself worked in all 74 stores tested.

What Shopify already does for you

All of this is on by default and needed no work from the merchants we tested:

Shopify's Spring '26 Edition says merchants are "UCP-enabled by default", and its changelog for the WebMCP checkout tools says they "don't expose a new API or require merchant configuration". The stores that didn't serve a profile were blocked, redirected or returned something else, not stores that had set something up wrong.

Which AI channels sell your products is also decided by Shopify and each channel. Per Shopify's help center, ChatGPT sends shoppers to your own checkout, while Microsoft Copilot, Meta and Google AI Mode and Gemini use Shopify-powered direct checkout for eligible US stores. You can switch individual channels off under Sales channels > Agentic, but you don't need to switch anything on.

What's still yours: the data agents ask for

The protocol is the same for every store. The answers it carries are not, and these are the ones that came back empty most often.

1. Return rules in admin, not only a refund policy page

When an agent asked "What is the return window for items bought from this store?", Shopify's policy tool returned a window in days for 33 of 176 stores that answered. For 141 stores, the only answer was Shopify's generic line that self-serve returns aren't available in customer accounts.

Almost all of those stores do have a refund policy: 167 of the 190 agents.md files link to one. The policy tool simply doesn't read that page. Of the 33 stores that did get a return window across, 20 answered in the generated wording Shopify builds from the return rules in admin. The other 13 answered with return or FAQ text the merchant had written for Shopify to use.

Fix: set your return window, fees and final-sale items in return rules under Settings > Policies. It is structured data Shopify can pass straight to an agent. Check it matches your refund policy page: one store we tested told agents both that it "does not accept returns" and that it accepts them within 30 days.

2. FAQ answers for shipping and cancellation

Agents asked two more questions shoppers ask before buying:

• "How much does shipping cost to Boston, and when will it arrive?" Only 4 of 176 stores returned anything. None gave a cost and a delivery time.
• "Can I cancel an order after placing it, and by when?" 8 of 176 returned an answer. All eight read like FAQ entries the merchant had written, such as "edit or cancel your order within 15 minutes of placing it".

The Search Engine Journal study found the same pattern on its 88 stores.

Fix: add short, number-led answers for shipping cost, delivery time and cancellation. Shopify's Catalog guidance points merchants to its Knowledge Base app for FAQs. Write rules as numbers ("within 24 hours", "free over $75") rather than "contact us".

3. Product data, starting with the category

Shopify Catalog decides which of your products AI channels can see. Its requirements include a title, at least one image and a price above zero, and they exclude uncategorized products. A product with no category in Shopify's product taxonomy doesn't reach AI channels through Catalog.

Beyond that, the fields Shopify says AI platforms consider are the ones a good Google Shopping feed already needs: title, description, images, type, vendor, collections and tags, barcode (GTIN, UPC or ISBN), and variants with clear option names. Barcodes are considered, not required.

Fix: filter your products for a missing category and fill it in. This is ordinary feed hygiene, not a new discipline.

4. A custom robots.txt, if you have one

64 of 197 stores block /policies/ in robots.txt, so crawlers can't read their refund, shipping and terms pages. Every one of the 64 serves an older-style robots.txt, which on Shopify means a custom robots.txt.liquid template. The 106 stores on Shopify's current default allow policy pages, and the default file now states that policy pages are crawlable.

This doesn't affect Shopify Catalog: Shopify says robots.txt rules only affect open-web discovery. It does stop search engines and AI crawlers that read the web from seeing your policies.

Fix: if your store has a robots.txt.liquid template, remove the Disallow: /policies/ lines, or delete the template if you no longer need it.

What you can skip

• A custom agents.md or llms.txt. Shopify's documentation calls the managed default "enough for most stores", and a custom template becomes yours to maintain. There is no evidence that llms.txt changes how AI engines treat a site.
• Declaring pickup to agents. All 183 UCP profiles we read declare shipping only, including stores whose profile offers location search. Google AI Mode and Gemini don't support pickup in direct checkout either. We found no merchant setting that changes this.
• Building anything for UCP or WebMCP. There is no plugin to install and no endpoint to configure.
• Product-page return and shipping markup, as an agent fix. Only 33 of 168 product pages carried return-policy structured data and 27 carried shipping details. Agents on Shopify get products through Catalog and UCP, not by reading that markup. It can still help Google's own shopping results, which is a separate decision.

How much does agentic shopping matter yet?

Less than the coverage suggests, but it is growing. Shopify reports that AI-driven traffic to its stores grew 197% year over year in the second quarter of 2026. It has not published how many orders are completed by agents. Traffic from AI is still small next to search: in one recent analysis of GA4 data from 100 sites, ChatGPT sent about 0.6% of the visits Google organic did.

That argues for doing the short list above, which also helps human shoppers and Google, and skipping anything sold as "agent optimization" beyond it.

Check your own store in two minutes

Replace yourstore.com with your domain:

# Is a UCP profile published?
curl -s https://yourstore.com/.well-known/ucp | head -c 300

# Does robots.txt block policy pages?
curl -s https://yourstore.com/robots.txt | grep -i "policies"

# What does an agent hear about your return window?
curl -s -X POST https://yourstore.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_shop_policies_and_faqs","arguments":{"query":"What is the return window?"}}}'

If the last command only returns the line about self-serve returns, set your return rules in admin and run it again. Swap the question for shipping or cancellation to test those answers too.

Getting your data into agents is half of it. Whether AI assistants recommend your store in the first place is a separate question, and it is the one generative engine optimization covers.