Agent-Buyable by Design: A Readiness Checklist for Agentic Commerce

What "agent-buyable" actually means

For thirty years the unit of online commerce has been a person and a page. A human looks at your storefront, reads the product copy, judges the reviews, fills the form, and pays. Agentic commerce changes the actor. An AI agent, working from a buyer’s stated intent, does the discovery, the comparison, and increasingly the checkout itself. The buyer says what they want; the agent finds it, confirms the terms, and completes the purchase.

That change has a structural requirement most site owners have never had to meet. A human can forgive a messy page, infer a price from context, and click a button labeled almost anything. An agent cannot. To be reachable at all, a business has to expose what the protocols call a machine-legible, policy-declared, verifiably-authorized transaction surface: a structured feed of what you sell and at what price, a declared set of rules for how a purchase may proceed, and a way to prove the buyer authorized the specific cart being bought.

This is the direct commercial analogue of a lesson that already governs search. A business can rank on page one and still be absent from the AI answer written above it. In the same way, a business can have a polished website and still be un-buyable by an agent that never sees a structured product or checkout surface. Being agent-buyable is not the same as having a good site. It is a separate readiness question, and it is now a concrete one.

The three protocols, in plain terms

The reason this is worth a checklist rather than a wait-and-see is that the infrastructure is shipped, not proposed. Three protocols went live within two weeks of each other in September 2025, each backed by named partners rather than a whitepaper. They overlap, they compete in places, and no single one has won. Understanding what each actually does is the precondition for preparing sensibly for all of them.

How to read this checklist

The checklist below is organized around the arc every one of these protocols has to complete, not around any single vendor’s spec, because betting your preparation on one protocol winning is exactly the mistake to avoid. Each item is a capability an agent needs from you, stated as something a small business can build or buy, followed by the protocol requirement it satisfies. The order is deliberate: the surfaces earlier in the arc, clean product data and reliable inventory, are the ones you need regardless of which protocol reaches you, and they are the ones worth building first.

One note belongs up front. The specs are real and the partners are named, but published adoption is concentrated among large platforms and their enterprise sellers, Etsy, Shopify, Walmart. There is not yet a measured body of evidence on how small independent merchants are being reached by agents in practice. This checklist is therefore framed as how to prepare for what has shipped, not a claim about traffic you are losing today.

Checklist part one: a machine-legible product surface

The first thing every protocol needs is a structured, current account of what you sell. An agent does not read your product page the way a shopper does; it reads a feed. Discovery, the step where an agent decides you are even a candidate, depends on your catalog existing as clean, structured data with the attributes an engine can match against a buyer’s stated intent: title, identifier, price, availability, variant, and the specifics that distinguish one option from another.

The concrete work here is mostly reconciliation, not invention. Most small merchants already hold this data, but they hold it in several places that disagree: a booking or point-of-sale system, a website catalog, a spreadsheet, a marketplace listing. If your price is right in one system and stale in another, an agent has no way to know which to trust, and a purchase built on the wrong record is exactly the failure the authorization protocols are designed to prevent. UCP’s premise, that an agent reaches into your backend for discovery and checkout, only works if that backend holds one correct version of each product.

• Maintain a single structured product feed with stable identifiers, current price, and live availability, reconciled from one source of truth rather than re-typed across tools.
• Include the attributes an agent matches on: variants, specifications, fulfillment options and any constraints that decide whether a buyer’s intent is a real fit.
• Make inventory reflect reality in near real time, so an agent is never offered, or worse, sold, something you cannot deliver.

Checklist part two: a policy-declared, authorizable checkout

Discovery gets an agent to your product. Completing a purchase requires a checkout an agent can operate under declared rules. This is where ACP’s Shared Payment Token and AP2’s mandate chain do their work, and where a merchant’s readiness is either present or absent. The token model lets the agent pay without holding the card; the mandate model records what the buyer authorized as a cart mandate so the transaction can be verified against it. Both assume your checkout can accept a purchase initiated programmatically and honor a declared set of terms: price, taxes, shipping, returns, and the conditions under which the sale is valid.

The practical readiness question is whether your commerce platform can expose an agent-facing checkout at all, and most small merchants get this capability through their platform rather than building it. That the two Google and OpenAI efforts both launched on Etsy and Shopify is the tell: for most independent businesses, being agent-buyable will arrive as a capability their existing platform turns on, which makes choosing and configuring that platform correctly the real lever, not writing protocol code.

• Confirm whether your commerce or booking platform supports an agent-initiated checkout (ACP Instant Checkout, or the equivalent) and enable it deliberately rather than by default.
• Declare your commercial policies in machine-readable form: price, tax, shipping, availability windows and returns, so the cart an agent assembles matches the cart you will actually honor.
• Ensure payments flow through a tokenized path where the agent never holds card data, which is the model both ACP and AP2 are built around.

Checklist part three: authorization and the human gate

The most important thing the protocols standardize is not payment, it is proof of authorization. AP2’s three mandates exist to answer a single question with cryptographic certainty: did the buyer authorize this exact cart at this exact price, or did the agent act beyond what it was given. For a merchant, honoring that chain is what separates a defensible sale from a disputed one. A purchase backed by a verifiable intent, cart and payment mandate is a purchase you can stand behind; one that is not is a liability waiting for a chargeback.

This is also where an agentic sale meets the same governance discipline as any other business AI. The authorization mandate is, structurally, a human-in-the-loop gate: it encodes that a person consented to a specific, bounded action before value moved. Preparing to be agent-buyable is therefore not only a data-plumbing exercise; it is deciding, in advance, which transactions may proceed on a standing authorization and which require a fresh human confirmation, and then declaring those boundaries so an agent respects them. The merchants who will do well here are the ones who treat the authorization surface as seriously as the product surface.

• Preserve the authorization record for every agent-mediated purchase (intent, cart and payment), so any sale can be traced to what the buyer actually approved.
• Set explicit boundaries for what an agent may complete unattended versus what needs a fresh human confirmation, and declare them rather than leaving them implied.
• Treat provenance as first-class: every record an agent acts on should carry where it came from and when it last changed, so a disputed order can be reconstructed instead of argued.

What is established, what is emerging, what is still marketing

What follows tiers the evidence, because the difference matters. What is established is that the protocols exist as shipped, documented specifications with named partners: ACP is a published Apache-2.0 standard live in ChatGPT, AP2 launched with more than sixty payment and platform partners, and UCP is in build with five named major retailers. None of that is forecast; it is a matter of public record and open repositories.

What is emerging is agentic commerce as an actual channel for independent merchants. The rails are laid and running for large platforms, but this is early infrastructure, and which protocol or combination becomes the default an agent uses to reach a small business is not settled. Preparing the surfaces that every protocol needs is the rational hedge precisely because the winner is not yet known.

What is still marketing, and should be named as such, is any confident number about revenue a small business is losing to agents today, or any claim that a specific storefront is already being shopped by AI at scale. The research pass behind this article found no measured evidence of small-merchant agentic-commerce adoption as distinct from the enterprise platform partners. This is readiness for a shift that has clearly begun, not a diagnosis of losses no one has yet measured.

Preparing without betting on a single protocol

The strategic move is not to implement ACP, AP2 or UCP directly; for almost every small business that is the wrong altitude and the wrong risk. The move is to build the surfaces all three depend on, because they converge on the same three requirements. Clean, reconciled, structured product data that reflects live inventory. A checkout your platform can expose to an agent under declared, honored policies. And an authorization and provenance record that lets you stand behind, or safely dispute, any sale an agent completes.

Those three surfaces are protocol-agnostic. They make you discoverable and buyable to whatever standard reaches you, and they are worth building on their own merits even if agentic commerce grew more slowly than its launch partners suggest, because a single trustworthy version of your product and transaction data is the foundation every other system you run depends on too. This is agent-buyable readiness: not a bet on which protocol wins, but the connective work that pays off under any of them, done in the order that puts the durable surfaces first.