ACP vs UCP: Which Agentic Commerce Protocol Should Merchants Prioritise in 2026?

Two protocols, two posture choices, one underlying truth: the merchants who win agentic commerce in 2026 are the ones who treat ACP and UCP as different distribution channels for the same well-governed catalogue.

OpenAI’s Agentic Commerce Protocol (ACP) and Google’s Universal Commerce Protocol (UCP) are the two open standards that matter for AI-initiated purchases right now. They share enough vocabulary that it’s tempting to treat them as interchangeable, and enough differences at the checkout and payment layer that picking one without understanding the trade-offs is a real risk.

This is a merchant-operations comparison. The goal is not to crown a winner, it’s to give you a sober decision framework so you can sequence feed, checkout, and payment work without blocking your own roadmap.

What ACP and UCP actually are

The two protocols were announced and shaped by different parts of the ecosystem, and that origin shows up in their specs.

ACP: OpenAI plus Stripe, distribution through ChatGPT

The Agentic Commerce Protocol is OpenAI’s open standard for letting ChatGPT surface, sell, and post-purchase manage merchant products. It was co-authored with Stripe, which is why the payment side reads natively as a delegated-payment flow rather than a checkout iframe.

The protocol breaks into three logical spec layers:

1. Product feed spec. A merchant-published catalogue, available as a daily file upload, an incremental API, or both in combination. The spec covers 60-plus product fields across identifiers, item information, media, pricing, availability, variants, fulfilment, merchant info, returns, performance signals, compliance, reviews, related products, and geo-tagging.
2. Agentic checkout spec. Defines how ChatGPT collects buyer intent and hands off a cart to a merchant-validated checkout session.
3. Delegated payment spec. Stripe-led primitive that lets a ChatGPT-initiated transaction settle through the merchant’s existing payment processor, with the merchant remaining the Merchant of Record.

Today, onboarding is gated. Merchants apply through chatgpt.com/merchants and the programme is described as “approved partners only”, which is the right framing rather than “anyone can ship.”

UCP: Google’s open standard for AI Mode and Gemini

Google’s Universal Commerce Protocol is positioned the same way at the surface level, an open standard for turning AI interactions into completed purchases, but its centre of gravity is different.

UCP sits on top of Google Merchant Center. If you’re already publishing a shopping feed for free listings or Performance Max, you have a head start on UCP. From there, Google offers two integration pathways:

1. Native checkout. The full agentic flow happens inside Google’s AI surfaces, AI Mode in Search, Gemini on web, using Google’s checkout primitives.
2. Embedded checkout. An iframe-based alternative for approved merchants with complex flows that don’t yet fit the native checkout model.

Critically, Google’s UCP page describes the protocol as interoperable with the Agent Payments Protocol (AP2), Agent2Agent (A2A), and Model Context Protocol (MCP). That choice positions UCP as a hub rather than a competing silo, and it’s why Shopify’s Agents documentation describes its MCP servers as UCP-compliant.

UCP is currently expanding to new verticals, Lodging and Food first, and operates a waitlist for general merchant access through Search Console.

Where Shopify lands

This is the most useful tie-breaker for many merchants. Shopify’s published Agents stack, Cart MCP, Checkout MCP, Order MCP, Global Catalog, Storefront Catalog, the Universal Cart API in preview, is explicitly UCP-compliant. If Shopify is your storefront, the protocol-level choice has already been partly made for you on the checkout side.

You can still feed ChatGPT through ACP from a Shopify storefront. The point is that the Shopify-platform tooling has chosen its protocol allegiance, and that lowers the cost of UCP-side checkout work for Shopify-based merchants.

Side-by-side comparison

The table flattens nuance, but the pattern is clear: ACP gives you a more prescriptive end-to-end stack from OpenAI plus Stripe, UCP gives you a more interoperable hub that lets other standards plug in.

The feed work is mostly shared

This is the part most merchant teams underestimate. The catalogue work for ACP and UCP overlaps far more than the marketing makes it sound.

A clean Merchant Center feed already covers the majority of ACP’s required fields, title, description, image URL, brand, GTIN, MPN, price, sale price, availability, condition, target country, shipping. The fields that genuinely differ are the AI-surface flags ACP introduces (is_eligible_search, is_eligible_checkout, is_eligible_ads) plus a handful of richer optional fields like q_and_a, structured reviews, popularity_score, pricing_trend, and related_product_id with relationship_type.

That means a feed-management programme that’s already solid for Merchant Center can be extended to ACP in days, not months, assuming your source of truth is well-modelled and your transformations are deterministic. If you’re still hand-editing CSVs every week, neither protocol is going to be kind to you, and the right next step is feed governance, not protocol integration.

For a deeper field-by-field walkthrough of the ACP feed, see our OpenAI ACP product feed spec walkthrough. For the Merchant Center side, our Universal Commerce Protocol guide covers the upstream work.

Where the two protocols actually diverge

Below the feed layer, the differences are real and operationally meaningful.

1. Checkout placement

ACP’s agentic checkout spec assumes the buying experience happens inside the AI surface, ChatGPT collects buyer intent and hands off a validated session for merchant fulfilment. UCP gives you a clearer choice: native checkout inside the AI surface, or embedded checkout via an iframe.

The embedded option matters for merchants with complex business logic, B2B approvals, tax exemptions, regulated categories, where shipping a full agentic checkout flow is genuinely risky on day one.

2. Payment routing and the role of Stripe

ACP’s delegated payment model uses Stripe as the primitive for routing a ChatGPT-initiated transaction back to the merchant’s existing payment infrastructure. Merchants stay the Merchant of Record, which means tax, refunds, and chargebacks remain on the merchant’s books, but the operational lift is Stripe-shaped.

UCP doesn’t have a Stripe-equivalent dependency. Payment routing lives inside Google’s checkout primitives. If you’re not already on Stripe and don’t want to be, that’s a real signal for which protocol to prioritise first.

3. Promotions

A small but practical difference: ACP exposes promotions only via the API path, not via the daily file upload. That nudges merchants who want sale-price coverage in ChatGPT toward the API-and-file hybrid integration pattern, which is the model OpenAI’s docs recommend anyway.

UCP inherits Google’s existing Merchant Center promotions surface, so the work you already do for free listings and Performance Max promotions carries through.

4. Interop with other agent protocols

Google’s explicit positioning of UCP as interoperable with AP2, A2A, and MCP is a structural choice. It means a Shopify Cart MCP server, a third-party agent framework, or a partner running on Anthropic’s MCP can plug into UCP without bespoke glue. ACP is more closely tied to OpenAI plus Stripe primitives, clean and prescriptive, but less of a hub.

If your roadmap involves Claude, Gemini, or other non-OpenAI agents acting on the same catalogue, UCP’s interop surface is the more natural foundation. If your roadmap centres on ChatGPT as the primary commerce surface, ACP’s tighter stack is the more direct path.

5. Partner access and operational cost

Both protocols are gated today, but the shape of the gating is different. ACP is a narrow approved-partner programme via chatgpt.com/merchants. UCP is a vertical-by-vertical expansion with a waitlist. The practical effect is that neither protocol is a “switch it on next week” project for most merchants, and the right preparation in both cases is feed hygiene, not protocol code.

A merchant decision framework

If you want a short answer rather than a table, here’s the framework most operations teams should run.

Step 1: Audit your existing distribution

Where does your shopping traffic come from today? If Google Shopping, Google Ads, Performance Max, and free listings drive most of your conversion, UCP readiness compounds with work you already do. If you’re already in the ChatGPT merchants application flow or piloting ACP, you have momentum, keep it.

Step 2: Decide your checkout posture

If you want the buyer to stay inside the AI surface, both protocols support that, UCP’s native checkout and ACP’s agentic checkout. If you need to keep the buyer on your own checkout for compliance, B2B approvals, or operational reasons, UCP’s embedded checkout is the cleaner story.

Step 3: Confirm your payment routing

Are you on Stripe today, and are you comfortable extending Stripe to the delegated payment role for ACP? If yes, ACP’s payment story is straightforward. If you’re on a non-Stripe processor and don’t want to re-platform, UCP avoids that dependency.

Step 4: Check Shopify and other platform alignment

If your storefront is Shopify, the Agents stack gives you Cart MCP, Checkout MCP, and Order MCP today, all UCP-compliant. That doesn’t preclude ACP, but it does mean UCP-side checkout work is mostly already done for you. For more, see our Shopify Agents stack explained post.

Step 5: Sequence feed before checkout

Whichever protocol you prioritise, ship feed readiness first. A clean ACP or UCP feed earns you discovery and citation signals well before either checkout integration is approved, and the feed work is the work that compounds across both protocols anyway.

Common mistakes to avoid

A few patterns to flag, drawn from teams already running through this decision:

• Treating the protocols as competitors and picking one at the catalogue layer. The catalogue is shared. Pick at the checkout and payment layer, not the feed layer.
• Onboarding checkout before the feed is spec-clean. Both protocols will surface feed problems hard when checkout goes live. Fix the feed first.
• Assuming Shopify’s UCP-compliance precludes ACP. It doesn’t. Shopify merchants can still submit ACP feeds; the Agents stack just gives you a head start on the UCP checkout side.
• Building bespoke transformations per protocol. Source-of-truth catalogue + deterministic transformations + per-channel validators is the durable pattern. Bespoke per-channel ETL is technical debt waiting to bite at the next spec update.

Where to go next

If you’re at the start of this work, the highest-leverage move is feed governance, a single source-of-truth catalogue, validated transformations, and per-channel readiness checks. Tools like our GTIN validator, product schema validator, and product feed auditor cover the universal hygiene work that pays back across ACP, UCP, and every other channel.

When you’re ready to integrate, use the protocol-specific guides, Agentic commerce shopping guide for ACP, Universal Commerce Protocol guide for UCP, and Universal Cart merchant feed impact for the cross-merchant cart angle, to scope the checkout and payment work without re-doing the catalogue.

The merchants who win 2026 don’t pick a side. They keep one well-governed catalogue and let each protocol earn its own checkout integration on merit.