Agentic commerce

Can an AI agent find your product, read the price and buy it?

Shopping agents read what your store publishes. Scan it the way an agent would.

Scan your store

Illustration. Not real data.

Scan your store

See what an agent would see.

Enter your store’s domain. The scan reads robots.txt, llms.txt, your homepage and, for a store, one product page.

No email. Runs in your browser. Nothing is stored.

How an agent shops

Four things an agent has to do.

Each one depends on something your store publishes. Here is what, and which Citehound check covers it today.

  1. Find

    The agent has to reach your catalog. It reads robots.txt and your sitemap, and looks for a UCP profile. UCP, the Universal Commerce Protocol, is a file at /.well-known/ucp that lists the services your store offers and where an agent reaches them.

    Checks: AI crawler access, Sitemap declared, UCP endpoint (stores only). How they work

  2. Read

    It needs your product’s name, price, availability and an image as structured data, not only drawn on the page.

    Check: Structured product data, read from one product page (stores only). How it works

  3. Compare

    To weigh your product against others it needs the same fields in a consistent form, and a plain statement of what your brand is and sells.

    Checks: Structured data (JSON-LD), llms.txt present. The commerce panel also reports whether the llms.txt is your own or a platform default. Methodology

  4. Buy

    It needs a path it can follow to complete the purchase, such as the cart and checkout capabilities a UCP profile declares.

    Check: the UCP endpoint confirms a profile exists and declares services. A check that follows a checkout path is in preparation. Methodology

The difference

Two stores. What can an agent read?

Same product, same price on the page. Only one of them publishes it in a form an agent can use.

Store A

An agent can act
Name
Canvas tote
Readable
Price
$42.00
Readable
Availability
In stock
Readable
Returns terms
30 days, free returns
Readable
UCP
Profile found
Present

Store B

An agent has to guess
Name
Canvas tote
Readable
Price
in an image
Not readable
Availability
Unknown
Not readable
Returns terms
Not found
Not readable
UCP
No profile
Absent

Illustration. Not real data.

What we check

Four things, and what a failure means.

4 of 8 commerce points

Stores only

UCP endpoint

The scan fetches /.well-known/ucp and passes it if it returns a valid merchant profile. A failure means an agent finds no profile, so it has no declared services to act on.

4 of 8 commerce points

Stores only

Product data

One product page is read for name, price, availability and an image in structured data, one point each. A failure means an agent has to guess the field from the page, or skip the product.

In the main score

Every scan

Trust pages

Contact signals and Author / about signals are checked today. A failure means nothing on the page tells an agent who runs the store. Shipping and returns pages are not checked yet.

Reported, not scored

Stores only

llms.txt authorship

The scan reads your llms.txt and reports whether it reads like your brand’s own file or a platform default. A platform default tells an agent nothing about your brand.

The commerce checks are scored separately, out of 8, and are never added to the main 100. How a scan decides it is looking at a store, and where that detection misses, is on the methodology page.

76%

of the DTC brand homepages we scanned had an llms.txt.

In the two we read by hand, the file was the platform’s default and said nothing about the brand.

Read the research report

Questions

Agentic commerce, briefly.

What is agentic commerce?

Agentic commerce is a shopping agent finding, evaluating and completing a purchase on a person’s behalf, inside a chat interface rather than a browser. It depends on machine-readable product data and a checkout path a program can complete, not just a webpage a person can read.

What is UCP?

UCP, the Universal Commerce Protocol, is a discovery standard that lets AI shopping agents find a store’s catalog, pricing and availability in a structured format. An agent looks for it at a fixed location, /.well-known/ucp, on a store’s domain. If the endpoint exists and is well-formed, an agent can act on it directly instead of relying on a general-purpose crawl of the storefront.

How do I make my store readable to AI shopping agents?

Start with the same foundation that makes any site readable to AI systems: confirm AI crawlers are not blocked in robots.txt, and that price, availability and product identifiers are marked up in structured data, not just rendered visually. Beyond that, a UCP endpoint at /.well-known/ucp gives an agent a direct, structured way to find your catalog instead of inferring it from a general crawl.

Does Shopify write my llms.txt?

Shopify serves a default llms.txt for the stores it hosts. In the two we read by hand, the file described the platform’s own commerce instructions and said nothing about the brand. Having the file is not the same as having said anything in it.

Do the commerce checks change my score out of 100?

No. The commerce checks are scored separately, out of 8, and are never added to the main 100. They run only when a scan treats your site as a store.

Your store

See what an agent would see.

Free, no email, nothing stored. You get the score, the checks that failed and the fix for each.