---
title: "Layers before protocols: ACP and UCP need good data"
description: "ACP and UCP standardize checkout, but both start from your product feed. A seven-layer model, what each protocol covers, and a bottom-up check list."
source_url: "https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/"
lang: "en"
---

[Insights](https://specoria.com/insights/) Infrastructure

# Layers before protocols: why ACP and UCP only pay off when your product data is right

ACP and UCP standardize checkout, but both start from your product feed. A seven-layer model, what each protocol covers, and a bottom-up check list.

By [Specoria Team](https://specoria.com/about/) October 21, 2026 8 min read

![Paper collage of stacked slabs: plaster hands and a stone arch at the top, blank cards in the middle, a heap of parcels and blank price tags at the base, and one violet paper key on the bottom edge.](https://specoria.com/insights/img/layers-before-protocols-acp-ucp-product-data-1-6fdc044f.jpg) Every floor of agent commerce rests on the catalog at the bottom, and no protocol can lift it higher. Illustration: Specoria · generated with AI

In this article

- [Seven layers, from the bottom up](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#seven-layers-from-the-bottom-up)
- [What ACP covers](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#what-acp-covers)
- [What UCP covers](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#what-ucp-covers)
- [Where both start: the bottom layer](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#where-both-start-the-bottom-layer)
- [How a lower-layer fault shows up higher](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#how-a-lower-layer-fault-shows-up-higher)
- [A bottom-up check list](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#a-bottom-up-check-list)
- [The bottom line](https://specoria.com/insights/layers-before-protocols-acp-ucp-product-data/#the-bottom-line)

When a store decides to "get ready for agents", the conversation often jumps straight to a protocol. Should we implement the Agentic Commerce Protocol (ACP) for ChatGPT, or the Universal Commerce Protocol (UCP) for Google? Both are documented, and both let an agent complete a purchase.

What gets less attention is what both protocols stand on. Neither one creates product data. Both start from the catalog you already have, and both pass along whatever is in it, right or wrong. In this article, we lay out a seven-layer model of agentic commerce, map what ACP and UCP actually cover, look at how readable retail pages are today, and end with a check list you can work through from the bottom up.

## Seven layers, from the bottom up

A useful way to picture agentic commerce is as a stack. Each layer depends on the one below it being correct.

- **Discovery and content.** Can an agent reach your pages and feeds, and read what is on them?
- **Identity and authorization.** Do you know which agent is asking, and what the shopper allowed it to do?
- **Catalog and cart.** Are products, variants, prices and stock accurate and the same everywhere?
- **Checkout.** Can a checkout be created, priced with tax and shipping, and completed?
- **Payment.** Can a payment credential pass from shopper to store without exposing card details?
- **Post-purchase.** Can the agent and the shopper follow the order, delivery and returns?
- **Observability and governance.** Can you see agent activity, measure it, and decide which agents may do what?

The rule that follows is simple: an upper layer cannot fix a lower one. A flawless checkout integration still fails if the variant the agent picked is out of stock. A payment token cannot correct the wrong price.

## What ACP covers

![Cut-paper stack leaning slightly, with a jagged crack running up from a torn blank price tag at the bottom to a crumpled receipt and blank envelope hanging from a stone arch, linked by one violet thread.](https://specoria.com/insights/img/layers-before-protocols-acp-ucp-product-data-2-9a162fd8.jpg) A small crack below quietly travels upward until the whole arrangement wobbles at the top. Illustration: Specoria · generated with AI

According to Stripe's documentation, ACP is "an open standard created by Stripe, OpenAI, and Meta" for how agents complete purchases on a buyer's behalf. Its building blocks are agentic checkout, cart and feed, delegated payment, delegated authentication with OAuth 2.0, and orders and webhooks for shipping, delivery and refund updates.

Mapped to the stack, that is layers 2 to 6: identity, cart, checkout, payment and post-purchase. Two details matter for the layers underneath.

- **It starts with a feed.** OpenAI's getting-started guide says: "Start your ACP integration by sharing a structured product feed with OpenAI." The feed has nine required fields: item ID, title, description, URL, brand, seller name, image URL, availability and price. Onboarding product feeds in ChatGPT is currently open to approved partners who apply.
- **Implementing it does not list you.** The ACP site answers the question directly: implementing the protocol does not mean your products will automatically be listed through AI agents, because "each AI platform will manage their own process."

The ACP site also says businesses can accept or decline transactions per agent or per transaction. That is a governance decision, and it belongs to layer 7.

## What UCP covers

Google announced UCP on January 11, 2026 as an open standard that works across the entire shopping journey, from discovery and buying to post-purchase support. It was co-developed with Shopify, Etsy, Wayfair, Target and Walmart. The UCP site lists its core capabilities as catalog search and lookup, cart building, identity linking, checkout and order management, with Agent Payments Protocol (AP2) support for payment.

So UCP reaches one layer lower than ACP: catalog search is part of the protocol. On Google, though, the catalog still comes from you. Google's UCP guide tells merchants to "use your existing Merchant Center account shopping feeds," and the Merchant Center integration page says only product listings with the `native_commerce` (checkout eligibility) attribute show the Buy now button. Checkout eligibility, in other words, is a field in your product data.

The same page sets the current scope: the UCP integration hub is rolling out gradually in the United States and applies only to products sold within the United States. If you sell in Turkey or Europe, UCP checkout on Google is something to prepare for, not something you can switch on today.

## Where both start: the bottom layer

If both protocols lean on your catalog and pages, the question becomes how readable those are. Adobe's figures from April 2026 give a rough picture for US retail. Its AI Content Visibility Checker scores how much of a page a machine can read. Across US retail sites, the average was 75% for homepages, 74% for category pages and 66% for product pages. For homepages, the best-performing sites scored 82.5% and the lowest 54.2%.

Those are US numbers from Adobe's own tool, and a page score is not the same as a feed's accuracy. Still, the pattern is telling: product pages, the ones a checkout protocol depends on most, were the least readable. Adobe's explanation is that retailers have thousands of SKUs and much of that content is currently invisible to language models.

## How a lower-layer fault shows up higher

Faults rarely show up where they start:

- **Stock (layer 3) becomes a failed checkout (layer 4).** The feed says a size is available, the agent builds a checkout, and the order fails at the last step.
- **Price (layer 3) becomes lost trust (layer 1).** The page, the feed and the structured data show different prices, and the agent cannot tell which is true. We looked at this in detail in [Your feed, your page and your structured data should agree on price](https://specoria.com/insights/feed-page-and-structured-data-should-agree-on-price/).
- **Policy text (layer 1) becomes an unanswered question (layer 6).** If return terms live only in a PDF or an image, an agent cannot tell the shopper whether a return is possible.
- **No logs (layer 7) means no verdict.** Without a way to see agent orders separately, you cannot tell whether a protocol project paid off.

## A bottom-up check list

Work through these in order. A "no" on a lower line is worth fixing before anything above it.

- **Discovery and content.** Product name, price, stock, shipping cost and return terms appear as text on the product page, not only in images or scripts. Your structured data matches the visible text, which Google's own AI features guidance asks for.
- **Identity and authorization.** You have decided which agents may read which pages, and if you link shopper accounts, permissions are narrow and can be revoked.
- **Catalog and cart.** Every variant has a stable ID that is never reused. Prices and stock agree across page, feed and structured data. You don't use placeholders such as "n/a" or "unknown" where a value is missing. Your feed refreshes at least daily; OpenAI recommends a full file once a day plus updates through the day.
- **Checkout.** A guest can complete checkout. Shipping cost appears before the final step. Repeating the same request does not create a second order.
- **Payment.** You know whether your payment provider supports agent payment tokens, and steps that need the shopper, such as 3-D Secure, hand back cleanly.
- **Post-purchase.** Order status, delivery estimates and return windows are available as plain facts, not buried in support pages.
- **Observability and governance.** Agent orders can be told apart in your reports, and someone owns the decision of which agents and protocols you accept.

Only when the first three lines are in good shape does it make sense to pick a protocol, and the choice then follows your buyers: ACP if they shop in ChatGPT and you are accepted to the program, UCP if they buy through Google's AI surfaces in a market where it is available. If you want the short version of why waiting for a single winning protocol is a mistake, we covered it in [Seven things people get wrong about agentic commerce](https://specoria.com/insights/seven-things-people-get-wrong-about-agentic-commerce/).

## The bottom line

ACP and UCP are worth understanding, and for some stores worth implementing. But they are upper layers. They carry your catalog to an agent and an order back to you; they do not make that catalog accurate. The stores that get the most from either protocol will be the ones that fixed the bottom of the stack first.

Specoria checks the lower layers an agent depends on: whether your product data is readable, whether page, feed and structured data agree, and whether a real browser agent can get from product page to checkout. Start with a [free readiness test](https://specoria.com/#contact).

### Sources

- Stripe, [Agentic Commerce Protocol](https://docs.stripe.com/agentic-commerce/protocols/acp), accessed October 20, 2026.
- Agentic Commerce Protocol, [agenticcommerce.dev](https://agenticcommerce.dev/), accessed October 20, 2026.
- OpenAI, [Agentic Commerce: Get Started](https://developers.openai.com/commerce/guides/get-started), accessed October 20, 2026.
- OpenAI, [Agentic Commerce: Products feed specification](https://developers.openai.com/commerce/specs/feed), accessed October 20, 2026.
- Google, [New tech and tools for retailers to succeed in an agentic shopping era](https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/), January 11, 2026.
- Universal Commerce Protocol, [ucp.dev](https://ucp.dev/), accessed October 20, 2026.
- Google for Developers, [Google Universal Commerce Protocol (UCP) Guide](https://developers.google.com/merchant/ucp), accessed October 20, 2026.
- Google for Developers, [About Universal Commerce Protocol (UCP) integration hub in Merchant Center](https://developers.google.com/merchant/ucp/guides), updated October 19, 2026.
- Adobe, [AI traffic grows but retail sites lag in AI search visibility](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable), April 16, 2026.
- Google Search Central, [AI Features and Your Website](https://developers.google.com/search/docs/appearance/ai-features), updated December 10, 2025.

## Related

- [A bot policy by page type: what to open to AI agents and what to keep closed](https://specoria.com/insights/bot-policy-by-page-type/)
- [When the agent checks the store: what the Chain of Trust demo means for your domain](https://specoria.com/insights/when-the-agent-checks-the-store-chain-of-trust/)
- [Shopify opened checkout to browser agents. What WebMCP means for every other store](https://specoria.com/insights/shopify-opens-checkout-to-browser-agents-webmcp/)
- [Agent protocol map: UCP, ACP, AP2, MCP, WebMCP](https://specoria.com/agent-protocols/) What each protocol does and what the free test checks
- [Agentic commerce statistics](https://specoria.com/agentic-commerce-statistics/) Sourced, dated figures you can cite

FREE TEST

## Is your store ready for shopping agents?

See in seconds how a shopping agent reads your store: free, instant, no email needed.

The free test runs in your browser and needs JavaScript. [Request the free deep audit by email instead →](https://specoria.com/#contact)

Your result opens on its own page. Want the full picture? The free deep audit and panel come next, by email.