What this check tests
The audit requests /.well-known/ucp once. The file is accepted only if the answer is a 200 with a JSON content type (application/json or a +json type) whose body is not the same as the body your site returns for a random made-up .well-known path. A catch-all page therefore cannot pass as a profile.
The body must be a JSON object with a top-level ucp object, and that object must have:
version: a string inYYYY-MM-DDform. Only the pattern is checked, not whether the date is a real UCP release.services: a non-empty object whose every value is an array.payment_handlers: a non-empty object whose every value is an array.
All three present gives pass, 100. Anything else fails with a message listing the problems, and there is no partial. capabilities, keys and the contents of the arrays are not read.
The check only applies to commerce sites. After the checks run, a result that is not a pass or partial is changed to skip with the message "Not applicable" unless the audit detected the site as ecommerce. Detection looks for signals such as a Shopify, WooCommerce, BigCommerce or Magento marker, Product JSON-LD, og:type of product, cart or checkout paths, price microdata and add-to-cart buttons, and needs a combined confidence of at least 0.5 that also beats any other site type the audit tries, such as local business. A valid profile on a store the detector missed keeps its pass. So a skip means "not detected as a store", not "no profile".
Why it matters to agents
UCP is the Universal Commerce Protocol, published at ucp.dev as an open protocol co-developed by large retailers, platforms and payment companies. It is an industry specification with dated releases, not an IETF or W3C standard. The specification says a business publishes its profile at /.well-known/ucp, that ucp.version, ucp.services and ucp.payment_handlers are required in a profile, and that platforms read it to negotiate capabilities such as catalog, cart and checkout before calling any endpoint.
Google's guide to generative AI search says protocols like UCP are emerging that will allow Search agents to do more. It does not say that publishing a profile changes how your pages are ranked or shown. The payoff is for stores that actually implement the protocol's services.
The audit is stricter than the specification in one place. The specification says the services and payment_handlers registries must be present even when empty, but this check needs both to be non-empty. It is also looser elsewhere: the specification forbids redirects on profile endpoints and requires a Cache-Control header with public and a max-age of at least 60 seconds. The audit tests neither rule.
How to fix it
Only if your store implements UCP services:
- Follow the specification's business profile example. An abridged skeleton:
{
"ucp": {
"version": "2026-08-25",
"services": {
"dev.ucp.shopping": [
{ "version": "2026-08-25", "transport": "rest", "endpoint": "https://example.com/ucp/v1" }
]
},
"payment_handlers": {
"com.example.processor": [{ "id": "processor", "version": "2026-08-25" }]
}
}
}
The real entries need the spec and schema members the specification lists; take them from its example.
- Serve it at
https://example.com/.well-known/ucpwithContent-Type: application/json, over HTTPS, without a redirect, and withCache-Control: public, max-age=3600or similar. - Use a dated release as the version. The specification says a business must not advertise a non-date string such as "draft".
- If your commerce platform provides UCP, check its documentation before hand-writing the file.
If you do not run a store, or do not offer UCP, the skip or fail is accurate and needs no action. Related: A2A agent card, since a UCP profile can list an A2A transport, and MCP endpoint, since it can list an MCP transport.