Three checkout tools. Three different pipes into your store. The question is which ones your customers are actually using—and whether your products are connected to any of them.

Shop Pay has been the standard answer to checkout friction for a few years now. 200 million users, a reported ~50% conversion lift over guest checkout, and baked directly into Shopify's ecosystem. It works. That part isn't in question.

What's new is that two more purchase surfaces just opened: Microsoft Copilot Checkout, announced at NRF 2026 in January, and ChatGPT's Instant Checkout, which launched in September 2025 powered by an open protocol called the Agentic Commerce Protocol. Both let shoppers complete purchases without ever leaving the AI interface they're in. That changes the conversion equation in ways most stores aren't ready for.

What is Microsoft Copilot Checkout and who is behind it?

Copilot Checkout lets shoppers buy directly inside a Copilot conversation. No redirect to your website, no new browser tab, no re-entering payment details. Microsoft announced it at NRF 2026 alongside a full suite of agentic retail tools, positioning Copilot as both a discovery and purchase surface for consumers and B2B buyers alike.

The integration runs through Stripe and the Agentic Commerce Protocol—the same open standard co-developed by Stripe and OpenAI. Launch partners included Shopify, PayPal, and Etsy. For Shopify merchants, the path to participation runs through your existing payment provider setup. The friction reduction is real. Redirecting a shopper from an AI chat to your product page, through your checkout, is a multi-step drop-off funnel. Copilot Checkout cuts most of those steps.

Microsoft's angle here is particularly worth watching for B2B sellers. The tools are designed to accelerate repeat purchases and reduce friction in order processing, not just one-time consumer buys.

How does ChatGPT Instant Checkout work differently from Shop Pay?

Shop Pay lives on your site. ChatGPT Instant Checkout lives inside ChatGPT.

When someone asks ChatGPT "what are the best running shoes under $100," it returns product results ranked on relevance—organic, unsponsored, based on what it finds across the web. If a product supports Instant Checkout, users see a "Buy" button right there in the conversation. They confirm shipping and payment details, and the purchase is done without visiting your store at all.

OpenAI launched Instant Checkout in September 2025, starting with Etsy sellers in the U.S. and announcing a rollout to Shopify merchants including Glossier, SKIMS, Spanx, and Vuori. The Agentic Commerce Protocol is open-source at agenticcommerce.dev, and merchants who want to participate can apply at chatgpt.com/merchants.

The fee structure is worth understanding before you sign up. OpenAI charges a 4% transaction fee on every completed Instant Checkout purchase, on top of standard payment processing (Stripe's roughly 2.9% + $0.30 per transaction). That's a real cost you need to run against your margins before activating.

I've been tracking this rollout across the stores we're auditing at WRKNG Digital. The merchants already appearing in ChatGPT product results share one thing: clean, complete product data. Structured titles, specific descriptions, populated attributes. The ones that aren't showing up have the same problem on the other end—sparse feeds, inconsistent data, missing fields.

Why do these new checkout surfaces create a different kind of conversion problem?

Traditional conversion rate testing is about your site. Page speed, button placement, checkout flow length. That work still matters for direct traffic.

But when a purchase originates inside Copilot or ChatGPT, you don't control the discovery experience. The AI decides what products to surface. The checkout happens in the AI's environment. Your store is essentially a backend—fulfilling the order, not driving the sale.

That's a fundamentally different problem. Getting your products into the AI's results is step one. Getting surfaced as a buyable product (with Instant Checkout enabled) rather than just a search result is step two. Completing the sale inside the AI interface is step three.

Most conversion tests are built around step three. Almost none address steps one and two. That's where the gap is, and it's widening fast.

Not great.

PayPal's adoption of the Agentic Commerce Protocol in October 2025 signals that this isn't an OpenAI-only experiment. The infrastructure is being built to support a multi-platform future where AI assistants become the primary purchase surface for a meaningful slice of ecommerce.

Which Shopify merchants can actually participate right now?

Here's the thing. Both Instant Checkout and Copilot Checkout are still in limited rollouts.

ChatGPT Instant Checkout is live for Etsy sellers and expanding to Shopify merchants through an application process. Microsoft's Copilot Checkout is similarly early-stage, with Shopify listed as a partner but without open enrollment for every store. Eligibility generally requires a supported payment provider (Stripe is the primary one), product data that meets AI-readable quality thresholds, and an integration with the Agentic Commerce Protocol either directly or through Shopify's platform layer.

The fact that the ACP specification is open-source matters more than it might seem. It means the ecosystem will expand quickly. More AI platforms, more checkout surfaces, more chances for your products to get purchased without the customer ever typing your URL. The merchants building toward ACP compatibility now won't need to scramble when the next platform adopts it.

What should your conversion testing strategy look like now?

Keep running your existing A/B tests on your site. Direct traffic still matters, and site conversion still matters.

But if Copilot and ChatGPT are becoming purchase surfaces—and the evidence says they are—the upstream variables in your conversion funnel now include things that most teams aren't measuring yet:

• Product title clarity. Can an AI understand exactly what this product is from the title alone, without needing to read the description?
• Description specificity. Materials, dimensions, use cases, compatibility. Vague copy doesn't surface well in AI results.
• Review count and recency. AI platforms use these as quality signals. A product with 4 recent reviews at 4.8 stars reads differently than one with 400 old reviews at the same rating.
• Feed completeness. Every required field populated, every variant mapped. Partial data means your product competes with one hand tied behind its back.

The conversion test is shifting upstream. It's moving from "did they click buy on my site" to "did my product get surfaced and purchased from an AI channel." Measurement tools for that aren't fully mature yet, but the stores tracking AI-channel purchase rates now will have a real data advantage in 12 to 18 months when this becomes mainstream volume.