Back to Blog
AEO / GEOLLM Advertising

The agentic commerce feed stack: what eCommerce teams actually need to track

July 29, 2026 7 min read

Everyone’s asking “which protocol do I need for AI shopping?” That’s the wrong question. “Product feed” for agents is really three questions stacked on top of each other — and once you see the layers, the whole thing gets a lot less scary.

If you sit anywhere on an eCommerce team, you’ve probably had a version of this week: a stakeholder forwards you an article about ChatGPT shopping, your agency pitches you an “AI SEO package,” and somewhere in your inbox there’s an acronym soup — ACP, UCP, AP2, MCP — with no clear answer on what your team is actually supposed to do on Monday morning.

Here’s the reframe that makes it manageable. The protocols everyone keeps conflating aren’t competing for the same job. They sit on different layers of the stack. Sort them by layer and the picture snaps into focus — and so does your to-do list.

The mental model

Three layers, not one feed

When people say “product feed for AI agents,” they’re smushing together three separate problems. Keep them separate and you’ll stop over-buying.

Layer 1 — Discovery: how the agent finds you. This is the actual catalog: getting your products into an agent so it can surface them in a conversation. It’s the closest thing to the “feed” you already know.

Layer 2 — Checkout: how the purchase gets executed. Whether the buy happens inside the AI surface or gets handed back to your own site. This is the layer that’s been the most turbulent in 2026.

Layer 3 — Payment & authorization: how consent is proven. How the agent proves the shopper actually agreed to spend the money, without exposing card details. Mostly your payments provider’s problem, not your feed’s.

A single agentic purchase can touch all three — discovered via one standard, checked out through another, authorized by a third. They stack. They don’t fight. Hold that thought, because it’s the whole argument.

The field guide

What you actually need to track

Here’s each standard worth knowing, sorted by how much it should occupy your roadmap right now — most actionable first.

ACP — Agentic Commerce Protocol

Backers: OpenAI + Stripe · Layer 1 & 2

This is the one with a concrete feed spec you can act on today. You push a product feed — JSONL, CSV, or XML, gzip-compressed — to an OpenAI endpoint, ideally refreshed daily. The required fields are the usual suspects:

  • Item ID, title (150-char cap), description (up to 5,000 chars)
  • Price with an ISO 4217 currency code, availability status
  • Image URL, brand
  • Eligibility flags controlling whether an item can be searched and whether it can be checked out

It’s Apache 2.0 licensed, and checkout runs on Stripe’s Shared Payment Token, so the agent never sees buyer credentials. The catch: OpenAI pulled back its native “Instant Checkout” in ChatGPT around March 2026 after underwhelming results, and pivoted toward a retailer-app integration model. But the ACP spec and feed are still live — so it remains the format to support if you want visibility inside ChatGPT.

Google Merchant Center feeds — the incumbent that quietly matters most

Backer: Google · Layer 1

This isn’t new and it isn’t sexy, but it’s now the substrate feeding Google’s AI surfaces and the emerging UCP standard. XML, TSV, or API submission; 170+ possible attributes, though the ones that carry the weight are title, description, price, availability, GTIN, product category, images, and shipping. If your Merchant Center feed is clean, you’re already halfway ready for agents — cleaning it up is arguably higher-leverage than chasing any single shiny new protocol.

UCP — Universal Commerce Protocol

Backers: Google-led coalition · Layers 1, 2 & beyond

The broad-coalition play: Google plus Shopify, Etsy, Walmart, Target, Wayfair, and a long list of payment and retail partners endorsing it. Announced in January 2026, self-serve since around June. Unlike ACP, it covers the full journey — discovery, cart, checkout, orders, and post-purchase — with a graceful hand-off to human checkout when needed. You publish a profile via a /.well-known/ucp JSON manifest, and it leans on your existing Merchant Center feed for the catalog itself. Transport is flexible (REST, MCP, Google’s A2A, or an embedded iframe). If you sell through Google Search’s AI Mode or Gemini and already run Merchant Center, this is the one to watch.

AP2 — Agent Payments Protocol

Backers: Google-initiated, now FIDO Alliance · Layer 3

Pure authorization — not a feed you build. It uses three cryptographically signed “mandates” — Intent, Cart, and Payment — expressed as W3C Verifiable Credentials, and it’s payment-method-agnostic across cards, bank transfer, and stablecoins. Moving to neutral FIDO Alliance governance (60+ orgs) makes it the lowest single-vendor-risk piece of the whole stack. You’d support it through your payments provider, not your catalog team.

MCP — Model Context Protocol

Backer: Anthropic · Transport layer

Worth understanding, but don’t file it as a rival feed standard. MCP is plumbing — a transport/context layer that other commerce standards (UCP included) use to expose merchant capabilities to agents. By raw developer adoption it dwarfs the commerce-specific standards, which is exactly why it’s likely to be the pipes underneath rather than a thing you “choose.”

The walled gardens — Amazon, Meta, Perplexity

Proprietary · Play by their rules

Amazon (Rufus, Alexa+, “Buy for Me”) is closed — it blocks external crawlers, keeps its own catalog structure, and sits out both ACP and UCP, so reaching it means playing inside Amazon. Meta is increasingly bundled into multi-standard adapters. Perplexity is a live shopping surface but without a well-documented open feed standard yet. Track them as channels, not as protocols to integrate.

Standard Layer What it solves Your move
ACP Discovery + checkout Selling inside ChatGPT Publish the feed if you’re on Stripe
Merchant Center Discovery Feeding Google’s AI surfaces Clean it up now — highest leverage
UCP Full journey Discovery + checkout across Google/Gemini Enable self-serve if on Shopify
AP2 Payment Proving user consent Support via your payments provider
MCP Transport The pipes underneath Nothing to buy — it’s plumbing

The bottom line

Support the layers, not the logos

The single most useful principle in this whole space: these protocols stack rather than compete, so stop trying to bet on a winner. Instead, get each layer ready and switch on integrations as each channel proves out. Concretely, whatever your team’s size:

  • Make your product data genuinely agent-ready first. A clean Merchant Center feed and solid Schema.org Product markup on your own pages benefit every protocol, whichever one wins. This is the durable investment.
  • Turn on ACP if you’re already on Stripe. Low effort, and it’s your ticket into the ChatGPT surface.
  • Enable UCP self-serve if you’re on Shopify / Merchant Center. It rides on infrastructure you already run.
  • Let your payments provider handle AP2. It’s their layer, not your catalog’s.

Bigger sellers increasingly route all of this through a single multi-standard adapter (Adyen Agentic and similar) to hit UCP + ACP + AP2 + Meta through one integration, rather than maintaining each by hand. If you’re scaling fast, it’s worth a look before you build four things separately.

Straight talk

One honest note before you over-invest

This space is moving fast, and the reporting on it is uneven. OpenAI’s checkout retreat in early 2026 is the clearest proof that “which protocol wins” is nowhere near settled — the leader in the category pulled back its flagship consumer feature after a few months. So treat feed-readiness as the thing you actually commit to, and treat the specific protocol integrations as switches you can flip relatively cheaply as each channel earns its place. Don’t torch a quarter’s roadmap building deep, bespoke plumbing for a standard that might get folded into another one by the time you ship it.

Get your data right. Stay flexible on everything above it. That’s the whole play.

Sources we used to research this piece

Related articles