In September 2025, OpenAI and Stripe launched Instant Checkout: buy a product without leaving ChatGPT. It was the headline demo for the entire agentic commerce category. By March 2026 it was gone, retired after roughly five months, with only around thirty Shopify merchants integrated and sales close to zero. Purchases completed inside the chat converted roughly three times worse than those redirected to merchant websites. Easy conclusion: the hype was empty. Wrong conclusion. The consumer button failed; the infrastructure beneath it kept getting built, and that infrastructure is what decides whether AI systems can recommend and transact with your business at all.
What actually happened
The flashy layer moved faster than the market was ready for. Consumers were not eager to complete purchases inside a chat window, and merchants had little reason to integrate for a trickle of transactions. OpenAI pivoted to retailer-operated experiences instead, the apps now used by companies including Walmart, Etsy, Target, and Instacart, where the agent recommends and the shopper completes the purchase on the merchant's own property.
Meanwhile the plumbing settled into layers rather than a single winner. The Agentic Commerce Protocol from OpenAI and Stripe survived its own failed demo as the standard for how an agent completes a checkout. Google and Shopify announced the Universal Commerce Protocol at the NRF retail conference in January 2026, covering the full journey from discovery through post-purchase, and Shopify made agent registration self-serve in June 2026. The payment networks moved too: Mastercard completed live agentic transactions in Asia-Pacific markets, Visa launched its Trusted Agent Protocol commercially, and American Express added purchase protection for registered AI-agent purchases in April 2026.
Read that list again and notice what it is: not a product launch, but rails. Rails take longer to matter and matter for much longer.
Where adoption actually sits
The useful numbers show a clear shape. An IBM Institute for Business Value study in 2026 found roughly 45% of consumers already using AI for at least part of the buying journey, and usage concentrates heavily at the start: around 62% for product comparison, against roughly 23% at checkout and 19% for post-purchase.
That distribution is the whole strategic point. People are not yet handing over their wallets. They are handing over the shortlist, which is the step that used to belong to search results and product pages. Whether you are on that shortlist is decided long before any checkout protocol becomes relevant to you.
Forecasts vary as forecasts do: McKinsey has estimated agentic AI could influence three to five trillion dollars in global retail commerce by 2030, and Morgan Stanley has predicted nearly half of online shoppers using AI agents by then. Treat those as direction rather than schedule, using the skepticism we set out in our guide to reading search statistics.
Why this matters even if you never sell a product online
Agentic commerce gets discussed as a retail story, and the discovery layer underneath it is not retail-specific. When someone asks an assistant to find a clinic, a contractor, a law firm, or an agency, the same sequence runs: parse the request, evaluate options from available data, return a shortlist. The transaction may happen by phone or in person, and the selection still happened inside the answer.
For service businesses that means the shortlist question is already live, while checkout protocols remain irrelevant. The work that gets you shortlisted is the work we describe in our AEO playbook, and you can measure where you stand today with our 15-minute self-test.
What to actually do now
The playbook is unglamorous, which is exactly why it will separate businesses that get chosen from those that do not:
- Make your data complete and true. For retailers that means accurate titles, attributes, pricing, and stock; for service businesses it means hours, service areas, prices or price ranges, and what you actually do. Agents cannot recommend what they cannot verify.
- Publish it in machine-readable form. Structured data and clean feeds beat beautifully designed pages that hide the facts in images, since agents query structure rather than admire layouts.
- Make your terms readable by machines. Returns, shipping, warranty, cancellation, and booking conditions buried in prose are effectively invisible, and they are precisely the constraints an agent filters on.
- Keep prices and availability accurate in real time. An agent that recommends you and finds a different price at the destination learns not to recommend you again.
- Test your own visibility. Ask the assistants to find a business in your category and see whether you appear. This costs nothing and most of your competitors have never done it.
- Fix your measurement. Sessions, bounce rate, and cost per click do not capture agent-influenced sales, and much AI-referred traffic arrives with no referrer at all. Watch branded search, direct traffic, and citation presence alongside the usual reports.
None of this requires a protocol integration for most businesses. It requires being legible, accurate, and trustworthy to software, which is the same thing being legible to search engines has always required, only stricter.
The lesson worth keeping
The Instant Checkout story is a useful inoculation against both hype and dismissal. Anyone who spent 2025 waiting for the winning checkout button wasted a year, and anyone who concluded in March that agent shopping was fake is now behind on rails that quietly went live around them.
The reliable move in both directions is the same: be the business the systems can understand and recommend. That work pays whether agentic commerce arrives at McKinsey's pace or half of it, and it is the identical work that earns AI citations today, which is the point of our SEO versus AEO comparison.
Frequently asked questions
Can AI agents actually buy things right now?
Increasingly yes, though the dominant pattern today is discovery plus redirect: the agent recommends and the shopper completes the purchase on the merchant's site. Fully autonomous checkout exists in specific integrations and is not yet how most transactions happen.
Do I need to integrate an agentic commerce protocol?
Most small and mid-sized businesses do not, yet. If you sell through a major platform, protocol support is arriving through the platform itself, which is how Shopify merchants gained coverage without building anything. Spend your effort on data quality and visibility first, because those decide whether any protocol ever gets a chance to matter for you.
How do I know if agents are already sending me business?
Look for AI assistant referrers in your analytics, and expect them to undercount badly, since a large share of AI-referred traffic arrives without referrer data. Rising direct traffic and branded searches alongside flat organic clicks is often agent influence showing up in disguise.
Is this just the voice search hype again?
A fair question, and the difference is where the money moved. Voice search never rewired payment networks or produced open protocols from Google, OpenAI, Stripe, Visa, Mastercard, and American Express. Infrastructure investment of that kind is a stronger signal than any forecast.
The bottom line
The first attempt at AI shopping failed publicly and the rails underneath it kept being laid, which is roughly how every infrastructure shift has looked from the inside. You do not need to bet on a timeline. You need your data accurate, your facts machine-readable, your terms explicit, and your visibility measured, because those are the requirements for being shortlisted today and transacted with later. If you want to know how your business currently appears to the systems already making those shortlists, our free SEO and AEO audit tells you exactly that, in plain language.
