Agentic Checkout Shipped, Then Stalled. The Real Numbers

Every forecast deck says agentic commerce is a trillion-dollar wave. The most-watched player in it just scrapped its flagship feature. Both of those are true, and the gap between them is the most useful thing I can show you this quarter.

“

"Agentic will be a paradigm shift for e-commerce," says Nathan Feather, who covers small- and mid-cap internet companies at Morgan Stanley Research.

So this is a research read, not a hot take - the real numbers, where they come from, what the most credible independent analysis says, and what our own audit found about whether mid-market stores are even ready.

What shipped - and what just got pulled

OpenAI launched "Buy it in ChatGPT" - Instant Checkout - in February 2026, built on the Agentic Commerce Protocol (ACP), an open standard co-developed with Stripe. The shopper completed the purchase inside ChatGPT, OpenAI charged the merchant a 4% fee, and ACP ran a delegated-payment flow where your product feed made the catalog legible, the agent assembled the cart, and merchant endpoints confirmed the order. Real infrastructure, published spec, working rails.

Then, in March, OpenAI walked it back. It scrapped native in-chat checkout and started routing buyers back to the retailer's own site instead. The official line from an OpenAI spokesperson:

"We are evolving our commerce strategy within ChatGPT to better meet merchants and users where they are. Instant checkout is transitioning to apps, where purchases can occur more seamlessly.

Read that plainly: the company most aggressively pushing in-chat checkout concluded the agent should hand the buyer back to the store. That is not a small detail. It is the entire thesis of this piece, confirmed by the one player with the most to gain from the opposite outcome.

Why it stalled - the numbers behind the retreat

The gap between "1M+ merchants coming" and reality was brutal: only around 30 Shopify merchants were ever live on Instant Checkout, out of millions.

“

"The biggest challenge is that for agentic technology to scale, it has to work for everyone. Today's digital commerce is fragmented—everyone is building custom plugs for custom sockets, which is slow and expensive. The future of commerce and payments should be open, collaborative, and built for everyone to succeed."

The reasons that surfaced in reporting are worth itemizing, because they're structural, not teething:

1. High-intent browsers, low in-app conversion. Users asked ChatGPT plenty of product questions - comparing laptops, hunting gifts - but didn't complete the purchase inside the app. The intent was real; the in-chat checkout wasn't where they wanted to close.
2. The sales-tax problem. As of February 2026 OpenAI had not built a system to collect and remit US state sales tax - a live regulatory blocker on every transaction passing through its checkout.
3. Onboarding friction and errors. Getting merchants integrated was arduous and the flow was error-prone, which is how you end up with ~30 live stores instead of a million.

And the merchant-side objections never went away either: a 4% cut on top of standard processing, no email or retargeting pixel captured on an in-chat sale, and wariness about feeding catalog and conversion data to a platform that could later pick winners. The Amazon lesson, learned in advance.

Where ChatGPT's product picks actually come from

This is where the data gets specific - and verifiable. A Semrush study, covered by Search Engine Land, decoded how ChatGPT assembles its shopping carousels by analyzing 43,000 carousel products across 10 verticals. The headline finding:

"ChatGPT sources 83% of its carousel products from Google Shopping via shopping query fan-outs.

The researchers even decoded base64 fields in ChatGPT's responses that exposed Google Shopping parameters like productid and offerid. And in a separate 100-run test, ChatGPT's top product appeared in Google Shopping's top 3 results 75% of the time. ChatGPT isn't building its own product index - it's reading Google's.

The takeaway is blunt: if you're an e-commerce brand, your Google Shopping feed needs to be the focus, and every listing has to be current.

The demand signal underneath is still very real

Do not read "checkout stalled" as "no opportunity." The traffic and intent are surging - it's only the in-chat transaction that retreated. Keep the measured rows and the forecast rows separate and the picture is clear:

The agent sends the buyer; your page closes them. A year ago that was my recommendation. Now it's literally OpenAI's product strategy.

Our audit: are mid-market brands even ready?

Here's the part I most wanted to check, because every "should I integrate" debate assumes you could. I went back through our 50+ brand audit and scored each against the real prerequisite - not a checkout integration, but a clean Google Shopping feed and machine-readable product data an agent can act on.

The finding that jumped out: the checkout fee was never most brands' problem, because they aren't discoverable in the first place. The recurring failures were the unglamorous ones:

• Product feeds missing GTINs, or with stale price and availability an agent can't trust
• Returns and shipping terms living in a PDF or a help page, not in structured machine-readable fields
• Schema injected client-side with JavaScript, so the agent's crawler never sees it - the same trap that breaks ChatGPT Shopping visibility
• No Merchant Center feed at all - which, as covered in the AI Mode piece, is the gate to the whole agentic surface on Google's side too, and by Semrush's data, to ChatGPT's picks as well

The bet that actually matters: the feed, not the checkout

OpenAI's retreat is the clearest possible lesson against betting on one assistant's checkout UX. Microsoft Copilot shipped its own checkout, Perplexity has Instant Buy with PayPal, and Google's agentic surfaces keep expanding - any of them could lead, stall, or pivot like OpenAI just did. The one constant across all of them is the product feed and clean structured data. That's the asset that doesn't get scrapped in a March product review.

So the asymmetry is stark. The cost of being ready is data hygiene you should be doing for AI search anyway. The cost of being late is sitting out a channel that, even on Morgan Stanley's conservative forecast, becomes a tenth of e-commerce - while your discoverable competitors get the recommendation every time.

What I'd actually do - 5 moves

1. Make the Google Shopping feed your #1 priority. The Semrush finding says it plainly - 83% of ChatGPT's carousel comes straight from Google Shopping - so fix the Merchant Center feed - GTINs, live price and inventory, complete attributes - before anything else. It's the gate to both Google's AI surfaces and ChatGPT's picks. See the Shopify schema guide.
2. Fix server-rendered structured data. If your schema is injected with JavaScript, agents can't read it. Confirm it's in the raw HTML source.
3. Make returns and shipping machine-readable. Structured fields an agent can parse without a human, not a PDF. This is also our most common audit failure.
4. Stay protocol-ready, not assistant-dependent. Don't build your strategy around one checkout that just proved it can be pulled. Clean data positions you for whichever model wins.
5. Track AI referrals now. Tag AI-source traffic and watch its conversion - see our AI-search KPI guide. The buyers are coming from the agents to your site; measure them.

Want to know whether the AI agents can even find and read your store - the prerequisite the data says drives ~83% of ChatGPT's picks? Run GEOlikeaPro's Visibility Vitals checker and start with the feed.