AI commerce readiness: six signals that shape how agents book, pay, and sign up

1. What changes when the buyer is a model

The mainstream analytics funnel assumes a human is on the other side of every click. Attention is scarce, copy earns trust, and CSS controls whether the CTA gets tapped.

None of that is true for an agent. An agent has patience, no visual bias, and one job: parse the page, find the transaction path, decide whether to trust it. If your <button> is a div, the agent walks away. If your card number field is autocomplete=off, the agent silently swaps you for a site that spelled it right.

The good news: none of this is philosophical. Every one of the six signals below is verifiable with a curl, a browser devtools panel, or a two-minute scan. If a machine can read the signal, a machine can grade you on it.

2. Signal 1: form semantics

A checkout is a form. That should be obvious, but many modern SaaS shells replace <form> with a stack of divs plus a JavaScript submit handler. Fine for a human clicking. Broken for anything that inspects DOM structure.

An agent that can recognize a <form> wrapping the credit-card block gets an anchor for everything else. Attributes hang off it: method, action, aria-label. Skip the form element and you force the agent to guess. Guesswork degrades trust. Trust degrades conversion. Every downstream signal in this list matters less if signal 1 is missing.

Fix in five minutes: wrap the payment section in <form>, add aria-label="Checkout", keep the submit handler you already have.

3. Signal 2: autocomplete hints

The autocomplete attribute was defined for password managers, but it turns into the fastest possible protocol for an agent to identify field intent. autocomplete="cc-number" is unambiguous. name="txt_field_7" is not.

The full canonical list is well documented (HTML spec section 4.10.18.7). For a checkout page you need at minimum:

• cc-name cardholder name
• cc-number the card number
• cc-exp or cc-exp-month + cc-exp-year
• cc-csc CVC
• billing address-line1, postal-code, country

For signup: email, username, new-password. That is it. Six attributes flip a page from opaque to legible.

4. Signal 3: provider fingerprint

Stripe Elements, Paddle Checkout, Lemon Squeezy, Creem, PayPal: each renders its widgets through a known iframe origin. An agent scanning the page for iframe[src*="stripe.com"] gets an instant answer: this is Stripe, I know the flow, I can complete it. No provider fingerprint, no confidence.

You are already loading these iframes. The signal is free. But it is only usable if the iframe is visible in the DOM at scan time, not lazy-loaded behind a "Show payment" toggle. Render the payment provider on the checkout page load, not on button hover.

5. Signal 4: declarative flow files

Robots.txt taught the web one lesson: agents prefer to read the machine-readable summary before crawling the whole site. The same pattern is emerging for commerce.

Two files matter:

1. /.well-known/agent-checkout.json a schema describing your checkout endpoints, supported currencies, tax handling, and refund policy. Draft spec at agentfix.pro/blog/llms-pay-spec.
2. llms-pay.md a short markdown file at the site root, referenced from llms.txt, telling the agent which endpoints to hit, in what order, and what to expect back.

Neither file is required by any standards body today. Neither adds runtime cost. Both give a scanning agent a single fetch that answers "can I buy here." If your competitors have them and you do not, agents pick the competitors.

6. Signal 5: sign-in options and SSO

Not every commerce path starts with cash. Bookings, quotes, and free trials all start with signup. An agent completing signup wants:

• a visible email field with autocomplete="email"
• a password field with autocomplete="new-password"
• SSO buttons with recognizable ARIA labels ("Sign in with Google", not "OAuth 3")
• optional passkey support (credential.get() in-page or a WebAuthn button)

The passkey signal is worth a paragraph on its own. Passkey-capable signup is one of the few commerce readiness checks that actively saves the agent work: no email verification loop, no password reuse question, and the identity is already verified by the platform. Sites shipping passkeys today are ranking themselves ahead of the competition without publishing a single blog post.

7. Signal 6: pricing metadata

Once the checkout works, agents still need to compare prices before they hit "buy. " The old answer was schema.org Product + Offer JSON-LD. That still works. The new answer is a companion llms-pay.md block listing SKUs, prices, tiers, and the direct checkout URL for each.

Machine-readable pricing lets an agent quote your site in a comparison table. Ambiguous pricing (image-only, JavaScript-gated) means the agent quotes someone else.

8. The 60-second self-check

Open your checkout page. Open devtools. Run:

document.querySelectorAll('form').length
document.querySelectorAll('[autocomplete^="cc-"]').length
document.querySelectorAll('iframe[src*="stripe"], iframe[src*="paddle"], iframe[src*="creem"]').length

You want all three greater than zero. If any returns zero, that is a five-minute fix and a five-percent conversion lift the next time an agent runs your funnel.

Then curl these:

curl -s https://your.site/.well-known/agent-checkout.json
curl -s https://your.site/llms-pay.md

Two 404s? Your site is invisible to shopping agents. Two 200s? You are ahead of every SaaS that shipped a landing page last week.

9. Key takeaways

10. What to do next

If you sell online, run the 60-second check above and fix the first zero you see. If you want a full grade across all six signals plus the 34 existing AI-readiness checks, scan your site at agentfix.pro. The scanner will tell you which signals are missing, why they matter, and what the fix looks like on your stack.

Commerce readiness is not a checklist you finish. It is a protocol you keep in sync with the agents that will decide, more each quarter, whether your funnel exists.