ACP: Which Merchants Really Use It (2026)

What Is the Agentic Commerce Protocol (ACP)?

Update, July 23, 2026. On March 24, 2026, OpenAI ended Instant Checkout, the product that carried ACP checkout inside ChatGPT. OpenAI’s own explanation: the initial version did not offer the level of flexibility it wanted, so merchants now use their own checkout experiences while OpenAI focuses on product discovery (OpenAI, March 24, 2026). ACP itself did not stop. It changed jobs, from checkout to discovery. Anything you read that still describes ACP as “the checkout inside ChatGPT” predates that change.

The Agentic Commerce Protocol (ACP) is a protocol co-developed by OpenAI and Stripe, announced in September 2025 alongside Instant Checkout (OpenAI, September 29, 2025). It is the technical layer that connects merchant data to ChatGPT: product feeds, promotions, availability, attributes. Until March 2026 it also carried an in-conversation checkout. Since then, its job in ChatGPT is product discovery, and the purchase completes on the merchant’s own store.

Which Merchants Actually Use ACP

This is the question most pages about ACP avoid, because the honest answer is uncomfortable: the list of merchants that ever completed a purchase through ACP is very short, and none of it is running today.

What circulates instead is a merged list of twenty or so logos that blends three very different things: merchants that had a live ACP checkout, merchants that only feed product discovery, and merchants that were named in a press release and never confirmed afterwards. Below, those three categories stay separate.

Category A: merchants with a demonstrable ACP checkout (all stopped)

That is the whole category. Two named merchants, plus a group of Shopify merchants that was never disclosed and whose size is only known through two conflicting estimates.

Category B: ACP for product discovery only (the only solid named list today)

This is the list to use if someone asks you for real ACP examples in 2026. It is short, it is named, and it comes from a primary source.

Category C: announced, with no proven deployment

Category C is where most published ACP merchant lists come from. A brand named in a launch post is a commercial intention, not a deployment. Treat “onboarding” and “coming soon” as what they are.

How to read the three categories

• Category A answers “who sold through ACP”. Answer: almost nobody, and nobody today.
• Category B answers “whose products ACP carries into ChatGPT”. Answer: seven named US retailers plus the Shopify merchant base.
• Category C answers “who said they would”. Answer: a lot of brands, with no follow-up evidence.

If a page hands you a single list of twenty logos, it has merged the three.

Why the List Is So Short

Three documented reasons, all from the same reporting on the March 2026 retreat.

1. Onboarding was heavy and the product was unreliable. Forrester analyst Emily Pfeiffer describes a laborious merchant onboarding process and an Instant Checkout that was prone to errors, with limited product selection six months after launch (CNBC, March 20, 2026).

2. Buying inside the chat converted worse than a redirect. Walmart measured a conversion rate three times lower when customers bought directly inside ChatGPT than when they were sent to Walmart’s own site. Those figures were first reported by Wired and relayed by CNBC (CNBC, March 20, 2026).

3. Volume never materialized. Etsy reported a relatively low purchase volume and characterized the technology as still nascent (CNBC, March 20, 2026).

None of this means AI shopping is small. A March 2026 Semrush survey of more than 1,000 US consumers, relayed by CNBC, found that 22% had already bought a product inside an AI tool. The gap is between buying because of an AI tool and buying inside the chat window through ACP. The first is growing. The second stopped.

What the ChatGPT Checkout Cost Merchants

In January 2026, Shopify confirmed to The Information that its merchants would pay 4% to OpenAI on sales made through the ChatGPT checkout, on top of Shopify fees. OpenAI never published that rate. The checkout in question stopped roughly six weeks later (The Information, January 21, 2026, corroborated by PYMNTS, January 21, 2026).

Two limits matter. That rate concerned Shopify merchants only: the terms agreed with Etsy and Walmart were never made public, and OpenAI declined to disclose additional financial details (CNBC, March 20, 2026). And the current Shopify documentation on agentic storefronts mentions no additional ChatGPT fee, since the payment now happens on the merchant’s own store (Shopify Help Center).

Two Different Things Are Called ACP

A naming trap worth flagging before you search further. This page covers the Agentic Commerce Protocol. The Agent Client Protocol, also abbreviated ACP, is an unrelated standard that connects AI agents to editors and command-line tools; GitHub Copilot CLI has supported it since January 2026 (GitHub Changelog, January 28, 2026). If a result talks about IDE sessions and stdio transport, it has nothing to do with commerce.

How ACP Works Today

Since March 2026, the ChatGPT journey looks like this:

User: "I'm looking for a 30L waterproof hiking backpack"
       |
       v
[1] ChatGPT interprets the purchase intent
       |
       v
[2] ACP carries merchant feeds and product data (Discovery)
       |
       v
[3] Products are compared side by side: price, reviews, features
       |
       v
[4] The user picks an option
       |
       v
[5] Checkout on the merchant's own store, in a ChatGPT in-app browser

The 4 ACP Capabilities

The specification still defines four capabilities. What changed in March 2026 is the product, not the protocol.

1. Discovery - The agent reads your catalog, your product pages and your feeds, and filters on attributes (size, color, price, availability). This is where ACP is actually deployed today.

2. Cart - The agent can assemble a cart, manage variants and apply promotions where the integration supports it.

3. Checkout - Still specified, no longer used inside ChatGPT. Since March 24, 2026, the buyer completes the purchase on the merchant’s own checkout (OpenAI). Other AI channels handle this differently, which is why the ACP versus UCP comparison below matters.

4. Post-Purchase - Order tracking, shipping information and returns are covered by the orders part of the specification. We have found no published deployment data for this capability.

Specification Status

The specification is hosted at github.com/agentic-commerce-protocol, maintained by OpenAI and Stripe, published under Apache 2.0, and still labelled beta. The last stable version is 2026-04-17 and covers cart, feed, orders, authentication and MCP. Earlier versions: 2025-09-29, 2025-12-12, 2026-01-16, 2026-01-30. A move to a neutral foundation has been announced, without a date.

Why ACP Still Matters

Before ACP, every AI platform built its own merchant integrations: fragmentation, slow adoption, uneven experience. ACP standardizes agentic commerce the way HTTP standardized the web (Shopify, 2026). The March 2026 retreat narrowed what the protocol does in ChatGPT, it did not remove the standardization. Merchant data quality is now the deciding factor for whether your products appear in a side-by-side comparison, which is exactly what OpenAI describes in its March 2026 announcement.

ACP vs UCP: What Is the Difference?

One geographic point often missed: Instant Checkout launched for US users buying from US Etsy sellers, and stopped before any announced expansion (OpenAI, September 29, 2025). The retailers integrated with ACP for discovery are US chains.

Which One Should You Choose?

Both, but for different reasons than in 2025. ACP connects you to ChatGPT product discovery. UCP connects you to the Google ecosystem (AI Mode, Gemini) where a direct in-channel checkout can still exist depending on your setup. The technical prerequisites overlap heavily: complete product data, coherent prices, real availability, accessible pages. Working on those serves both, whatever each platform decides to do with checkout next.

How to Get Your Shopify Store Ready

The work is no longer about switching on a checkout. It is about being correctly represented in discovery, and being easy to buy from once the shopper lands on your store.

Step 1: Check How Your Store Is Exposed to AI Channels

Exact path: Shopify admin, then Sales channels, then Agentic.

According to Shopify documentation, agentic storefronts are active by default for eligible stores, and products reach AI channels through Shopify Catalog. For ChatGPT specifically, Shopify describes it as a discovery-focused referrer: the shopper completes the purchase on your own store checkout, in a ChatGPT in-app browser or in a new tab. Orders show up in your admin with channel or referrer attribution.

Step 2: Make Sure Your Checkout Survives an In-App Browser

Since the purchase happens on your store, inside an embedded browser, the usual friction points become AI commerce problems: an interstitial popup, a cookie wall that blocks the page, a login requirement, a slow first render. Test your product page and checkout in an embedded browser context, not just in your desktop Chrome.

Step 3: Verify Your schema.org Product

Product data is what discovery consumes. Each product page should contain at minimum:

• @type: Product with name, description, image, sku
• offers with price, priceCurrency, availability, url
• brand with name
• aggregateRating if you have reviews (strongly recommended)
• shippingDetails with delivery time and cost
• hasMerchantReturnPolicy with return policy details

On Shopify, most modern themes generate basic schema.org Product markup. But shippingDetails and hasMerchantReturnPolicy are often missing, and need to be added manually or via an app.

Step 4: Ensure Price Consistency

Products are shown side by side with price, reviews and features. Any inconsistency between your sources creates a distrust signal:

• The price in your schema.org offers.price must exactly match the price displayed on the page
• Strikethrough prices (compareAtPrice) must be higher than the current price
• The currency (priceCurrency) must match your target market (EUR for France, GBP for the UK)
• OG prices (og:price:amount) must be synchronized with schema.org

Step 5: Verify Real-Time Availability

An AI assistant recommending an out-of-stock product destroys trust immediately. Make sure:

• Shopify inventory is synced in real time (no update delay greater than 15 minutes)
• The schema.org availability field reflects actual stock (InStock, OutOfStock, PreOrder)
• Sold-out variants are correctly marked

The 5 ACP Readiness Criteria

1. Product Discovery (Complete Schema)

The agent must understand your products unambiguously. This requires a complete schema.org Product with all fields listed in Step 3. Incomplete product data limits what ChatGPT can put in a side-by-side comparison, which is where the decision now happens.

2. Price Accuracy (Consistent Prices)

Displayed price, schema.org price and OG price should be identical. A divergence between your own sources is the easiest signal to detect and the hardest to defend.

3. Checkout Handoff Quality (Post-Click Experience)

Since the purchase completes on your store, the handoff is the conversion moment: a checkout that works in an embedded browser, a structured return policy in schema.org, visible delivery terms. Without hasMerchantReturnPolicy, an assistant cannot state your return conditions before the shopper leaves the conversation.

4. Fulfillment Info (Structured Shipping)

Delivery time and cost should be readable before the click. The schema.org shippingDetails field is the machine-readable version of that. Include at minimum: shipping zone, estimated time, standard cost, express cost.

5. Trust Signals (Reviews and Certifications)

Reviews (aggregateRating), certifications (organic, vegan, etc.), years in business and order volume are the signals a shopper compares across the options presented. The more structured they are, the more likely they are to be carried into the comparison rather than lost in page copy.

How Verity Score Measures Your ACP Readiness

Verity Score automatically audits your ACP compatibility and assigns an ACP Readiness score out of 100, broken down into 5 dimensions:

Score Interpretation

• 80-100: your product data is ready for AI discovery.
• 60-79: usable but with gaps. Some comparisons may skip you in favor of better-structured competitors.
• 40-59: not ready. Major corrections needed on schema.org and product data coherence.
• 0-39: an AI assistant cannot reliably describe your products.

A readiness score is a proxy: catalog quality, structured data, policies, availability, pricing, feeds and OpenAI crawler access. It is not an eligibility verdict, and Verity Score does not claim to grant or deny access to any program.

Merchant Checklist: 8 Actions

1. Check your Agentic sales channel - Admin, then Sales channels, then Agentic. Confirm which AI channels your products are exposed to.

2. Test your product page in an embedded browser - This is where ChatGPT shoppers land. Popups, cookie walls and slow renders cost you the sale after discovery worked.

3. Enrich schema.org Product - Verify each product page contains: name, description, image, sku, brand, offers (price, currency, availability), aggregateRating. Use an app like JSON-LD for SEO or Smart SEO if your theme does not generate them natively.

4. Add shippingDetails to schema - Integrate shipping information into schema.org: zone, delivery time, cost. On Shopify, this typically requires a Liquid theme modification or a dedicated app.

5. Add hasMerchantReturnPolicy - Structure your return policy in schema.org with: return type (refund, exchange), timeframe (14, 30 days), conditions. Link it to your return policy page.

6. Synchronize prices - Verify that schema.org price = displayed price = og:price:amount. Watch currencies: a French store must use EUR, not USD.

7. Verify real-time inventory - Admin, then Products, then Inventory. Enable inventory tracking for each product. Verify that sold-out variants automatically switch to OutOfStock in schema.

8. Run a Verity Score audit - Enter your store URL and get your ACP Readiness score. Fix the identified gaps and re-run the audit to validate.

Correction Priorities

If your score is below 60, focus in this order:

1. Basic schema.org Product (name, price, availability) - immediate impact
2. Post-click experience in an embedded browser - where the sale is now won or lost
3. ShippingDetails + ReturnPolicy - strong differentiator
4. AggregateRating - comparison fuel
5. Cross-source price consistency - avoids distrust signals

Related articles

• GEO Audit: Complete Guide - The 9 factors that determine your AI visibility
• Sell on ChatGPT - Agentic Storefronts and Shopify optimization for ChatGPT Shopping
• UCP: Universal Commerce Protocol - The Google + Shopify protocol for AI agents
• Perplexity Shopping - Merchant guide and readiness score
• Schema.org for Shopify - Technical guide to Product and Offer structured data
• Pricing and AI Visibility - How agents compare your prices and promotions
• Shipping & Returns - Critical trust signals for AI agents
• Robots.txt and AI Crawlers - Configure access for GPTBot, ClaudeBot and PerplexityBot
• AI Buyer Score - How AI agents evaluate and rank your products

Ready to check your store? Run a free GEO audit →

The short version: ACP is alive, its checkout is not, and the merchant list that matters in 2026 is a discovery list. Work on the product data that feeds that list, and on the store experience that receives the click.