OpenAI quietly shut down Instant Checkout in March 2026. Fewer than 30 merchants had bothered to launch on it. For a program that was supposed to be the first real proof that AI agents would buy things on your behalf, that's not a soft failure, it's a rejection.
Here's the part that should actually worry retailers though: while the checkout bet was dying, AI-referred shopping traffic was exploding. Adobe measured 393% year over year growth in AI-referred visits to US retail sites in the first quarter of 2026, and that traffic started converting 42% better than it had months earlier. Salesforce put AI and agent influence behind roughly a fifth of 2025 holiday retail sales, something like $262 billion. Those two facts sitting next to each other tell you where the real fight moved.
Checkout was never the hard part
The idea behind Instant Checkout made sense on paper: let an agent find the product, add it to a cart, and pay, all without a human clicking through a funnel. It didn't stick, and not because the technology couldn't do it. Merchants didn't want to hand over the transaction, the margin, or the customer relationship to someone else's agent. What survived is a narrower model: agent for discovery, redirect to the merchant's own checkout. Google's Shopping AI Mode will add items to a cart and pay with Google Pay when a shopper says "buy for me," and Amazon is testing agents that pull products from other sites into an Amazon cart. Both keep the actual purchase inside a platform's own walls. Nobody wants to be the store that got disintermediated at the one moment money changes hands.
The traffic came from somewhere else
So if agents aren't closing the sale, what are they doing? Deciding what gets recommended in the first place. Salesforce found AI-referred shoppers converting roughly nine times more often than shoppers arriving from social media, which means the recommendation itself is doing most of the selling before a human ever lands on the page. That's a different kind of pressure than a checkout integration. It means your product page has to win an argument it can't see happening, inside a model that's comparing your delivery time, your return policy, and your price against three competitors at once.
Most stores are still speaking the wrong language
This is the part nobody's built for yet. One measure of AI visibility, how legible a page actually is to a model, put the average retail product page at about 66%. That's a third of the page's useful information effectively invisible to the systems now doing the recommending. Schema markup, structured delivery promises, machine-readable return windows and shipping fees: these used to be nice-to-haves for SEO. They're now the difference between showing up in an agent's shortlist and not showing up at all. Cross-border shoppers already rate clear delivery-fee information as essential before they'll buy, and that bar only gets stricter as agents do more of the comparing.
If you're running a Shopify store or building the stack behind one, the practical move isn't chasing an agentic checkout integration that even OpenAI couldn't make merchants want. It's auditing whether your catalog, your delivery promises, and your policies are actually readable by the systems sending you traffic. That's unglamorous work, structured data, consistent product descriptions, API access to real-time inventory and shipping, but it's the work that decides whether you get recommended at all.
We're here to help founders and teams design and build digital products that are built to scale with you, not slow you down. If you're looking to build something, get in contact with us today!