Hype Index: 'ChatGPT Doesn't Just Help You Find What to Buy, It Also Helps You Buy It'

TL;DR: OpenAI shipped Instant Checkout on September 29, 2025 and pulled it out of ChatGPT in early March 2026, a five-month run. On August 15, 2026 I took all 33 merchants named by name in six agentic-commerce launch announcements and checked each domain for a live /.well-known/ucp manifest, the file an agent has to read before it can transact. Eight serve one. Fifteen conclusively serve nothing. Ten blocked my request outright. All eight that work are Shopify stores, which means the merchants did not become agent-ready. Their platform did it for them.

The claim in the title is OpenAI's, word for word, from September 29, 2025. Etsy went live that day. Glossier, SKIMS, Spanx and Vuori were named as "coming soon" across "over a million Shopify merchants." Walmart and Sam's Club followed on October 14. Then Google, then Microsoft. Eleven months of launch announcements naming specific brands.

The nice thing about agentic commerce is that it is falsifiable. Either a machine can reach a merchant's checkout without a human, or it cannot, and the same test one step earlier asks whether an agent can even read the price.

What did I actually count?

Google's Universal Commerce Protocol specifies that "the business's profile at /.well-known/ucp describes capabilities for the protocol version it declares." It is the agent equivalent of robots.txt: one JSON file at a fixed path that tells any agent which checkout, cart and catalog capabilities the merchant supports. No manifest, no permissionless agent checkout.

So I built the list. Six announcements: OpenAI's launch (Sept 29, 2025), Walmart's partnership post (Oct 14), Google's holiday agentic checkout (Nov 13), Microsoft's Copilot Checkout (Jan 8, 2026), Google's UCP announcement (Jan 11), and Google Marketing Live (May 20). Every merchant named by name, deduplicated: 33. Then one GET per domain, apex and www, on August 15, 2026.

Which merchants are actually agent-ready?

Eight returned a valid UCP manifest: Glossier, SKIMS, Spanx, Reebok, e.l.f. Cosmetics, Rugs USA, Fenty Beauty and Steve Madden.

Every one of those manifests is about 4.1KB and points its MCP endpoint at a myshopify.com host. Glossier's resolves to glossier-admin.myshopify.com. Reebok's to abg-reebok-store.myshopify.com. This is not eight merchants that built agentic commerce. This is Shopify flipping a switch across its base, which is exactly what Shopify said it would do.

Fifteen conclusively served nothing: Etsy, Vuori, Quince, Walmart, Wayfair, Target, The Home Depot, Michaels, Poshmark, Ashley Furniture, Nike, Sephora, Ulta Beauty, Flipkart and Zalando. Zalando returned a 200 with an HTML page, which is a soft 404, not a manifest.

Ten I could not resolve, and I am reporting that as a failure rather than a finding: Sam's Club (400), Chewy, Urban Outfitters, Anthropologie, Macy's, Petco and Samsonite (403), Lowe's (500), Best Buy (301 to nowhere) and Alexander Del Rossa (connection failed). A bot wall that returns 403 to a plain GET is a fact about the merchant's edge, not proof of absence, and I did not probe further.

The other number I expected to find was dead press releases. There are none. All six announcements still return 200, and OpenAI's own help page on shopping in ChatGPT, updated two days before I ran this, still tells users that "for some eligible products and merchants, ChatGPT may also show an Instant Checkout option." Nothing was retracted. The page just quietly stopped describing what happens.

Why did the buy button come out of ChatGPT?

Because almost nobody used it, and the people who did converted worse.

Shopify president Harley Finkelstein put the number of merchants actually live at about a dozen when he spoke at Morgan Stanley's tech conference in March 2026, as reported at the time. Forrester principal analyst Emily Pfeiffer told CNBC roughly 30 Shopify merchants were available through Instant Checkout as of February. Those two numbers disagree by a factor of two and a half, and neither can be independently checked, because the Agentic Commerce Protocol has no discovery mechanism. Its own FAQ still says OpenAI and Stripe are "working to create discovery mechanisms for AI platforms to identify businesses that have implemented ACP." You cannot audit a protocol that nobody can enumerate, and where a discovery file does exist, as with the A2A agent card 50 named supporters were supposed to publish, the count comes back at one.

Walmart made about 200,000 products available and then measured the result. Daniel Danker, Walmart's EVP for AI acceleration, product and design, told WIRED that in-chat purchases converted at one-third the rate of shoppers who clicked out to Walmart's own site, and called the experience "unsatisfying." CNBC confirmed the same figure with Walmart. At the same March 4 conference, Danker described Instant Checkout as "a very temporary moment in time" and said "by this time next month, you will not see that experience anymore." He was right.

The feature also never grew past its launch scope. OpenAI's announcement promised multi-item carts "next." Eleven months later CNBC noted multi-item carts and loyalty linking as things Google had shipped and "two features OpenAI hasn't yet fully cracked." This is the same pattern as every agent demo that works until you give it real work: the demo bought one thing, and one thing was all it ever bought.

Is agentic commerce dead, then?

No, and anyone selling you that headline is doing the same thing the launch announcements did, in the other direction.

Every serious piece of infrastructure shipped after the retreat. Visa launched Intelligent Commerce Connect, a single on-ramp for agent payments, on April 9, 2026. American Express shipped its ACE Developer Kit plus Agent Purchase Protection on April 14, committing to cover charges caused by a registered agent's error. Google donated AP2 to the FIDO Alliance in late April. Google's UCP checkout is rolling out across Search and Gemini this summer with Nike, Sephora, Target, Ulta and Walmart named, and UCP's spec already includes a discounts extension that ACP's launch scope never had.

That is what my count is measuring and what it is not. The rails are real and getting realer. The consumer surface, the part where a chatbot completes your purchase, is eight Shopify stores out of 33 named brands, and eleven months after the first announcement it is still the platform doing the work.

What would change my mind?

A sweep like this one returning 20 of 33 instead of 8, with at least half of them off Shopify. That would mean enterprise retailers actually built something rather than endorsing a protocol in a press release. I will rerun it.

Until then, the honest read is the one Pfeiffer gave CNBC: "Everyone thinks everyone else has this figured out, or is farther ahead of them. The fact is that no one has this figured out." That is the same shape as the Gartner forecast that 40% of agentic AI projects get canceled, and it rhymes with the other thing nobody has figured out either, which is whether an agent that browses the open web can be trusted with your credit card at all.

Check the manifest before you believe the announcement. It takes one request.