Why AI Checkout Stalled: Discover in AI, Buy on Site

The most important agentic commerce story of 2026 is not that AI can buy things for you — it is that in-chat AI checkout quietly stalled, and the industry regrouped around a different model: discover in AI, buy on your own site. OpenAI launched Instant Checkout inside ChatGPT in September 2025, then walked it back by early 2026. The reason was not philosophical. It was conversion math, and it reshaped what every ecommerce merchant should build this year.

The retreat matters because it settles a question merchants have been asking since agents learned to shop: do you hand the transaction to the assistant, or keep it on your own storefront? The numbers that leaked out of Walmart, the figures Shopify published, and an independent consumer survey all point the same way. Shoppers are happy to discover products through AI. They are not, yet, finishing the purchase inside the chat window.

This guide walks the full ten-month arc — the September 2025 launch, the March 2026 retreat, the Universal Commerce Protocol, Shopify’s default-on Agentic Storefronts, and Amazon’s opposite bet — and turns it into a concrete merchant playbook. Every vendor-reported figure is labelled as such; the independent numbers are kept separate so you can see exactly where the hype ends and the evidence begins.

On September 29, 2025, OpenAI launched Instant Checkout inside ChatGPT: U.S. shoppers could buy directly from Etsy sellers in the chat, with “over a million” Shopify merchants — Glossier, SKIMS, Spanx, Vuori — promised as “coming soon.” It ran on the Agentic Commerce Protocol (ACP), open-sourced with Stripe, and rode a reported 700M+ weekly ChatGPT users (OpenAI’s own figure, dated to that announcement). Shopify President Harley Finkelstein called agentic commerce the “new frontier” for online retail.

Six months later, the frontier had receded. In the second week of March 2026, The Information reported OpenAI was scaling back its shopping plans — a scoop picked up by both CNBC and Modern Retail. By February 2026, only roughly 30 Shopify merchants were actually live on Instant Checkout, per Forrester principal analyst Emily Pfeiffer — a rounding error against the “over a million” once promised. Gartner analyst Bob Hetu put it plainly: “OpenAI underestimated how difficult the enablement of transactions was going to be.” For the narrower, day-of news, see our earlier look at OpenAI’s March retreat from in-chat checkout.

The clearest evidence came from Walmart, which made roughly 200,000 products available through Instant Checkout starting in November 2025. According to Walmart EVP Daniel Danker — first reported by Wired, then corroborated independently by CNBC and Search Engine Land — checkout completed inside ChatGPT converted at about one-third the rate of sending the same shoppers to walmart.com to finish the purchase. Roughly 3× worse, on real traffic, at scale.

Here is the twist that makes “kill it” the wrong conclusion: the same Walmart traffic showed ChatGPT driving roughly 2× the new-customer acquisition rate Walmart sees from search engines. So the assistant was a genuinely strong discovery channel — it just could not close the transaction as well as the retailer’s own checkout. Danker called the in-chat experience “a very temporary moment in time” at Morgan Stanley’s TMT conference on March 4, 2026, and said Walmart’s own Sparky assistant would instead “travel directly into ChatGPT and Gemini” — the merchant’s brain riding the assistant’s reach, rather than surrendering the checkout.

Independent data confirms the split. A Semrush-commissioned survey of 1,030 U.S. shoppers with AI experience, fielded in December 2025, found 50% had bought something after using AI to research — but only 22% had ever completed a purchase directly inside an AI tool. Meanwhile 69% expect AI to play a bigger role in how they shop. Direction of travel is up; today’s in-chat conversion is not there yet. The table below keeps the independently reported numbers separate from the vendor-stated ones so the pattern is legible either way.

Read as isolated headlines, the events of the last year look contradictory: OpenAI retreats, Google and Shopify launch a new standard, Shopify turns everything on by default, Amazon relaunches its shopping agent. Strung into one timeline, they point the same way. Nobody durable is betting on native in-chat checkout as the primary transaction layer. They are building rails for discovery through AI while keeping the purchase where the customer relationship lives.

If scraping is what broke in-chat commerce, structured feeds are what fix the discovery half of the equation. Google and Shopify launched the Universal Commerce Protocol (UCP) on January 11, 2026 as an open standard so agents share “one shared language” to find and transact with any merchant — endorsed at launch by 20+ retailers and platforms. On March 19, 2026 Google shipped UCP updates: multi-item carts, a real-time Catalog capability so agents pull live inventory and pricing instead of stale scraped data, and loyalty portability. We cover the mechanics in Google’s UCP multi-item carts and real-time Catalog, and how UCP sits against the rival standards in UCP vs ACP vs AP2.

Adoption followed fast. Shopify activated Agentic Storefronts by default for eligible merchants on March 24, 2026, making products from 5.6 million stores discoverable across ChatGPT, Microsoft Copilot, Google AI Mode, and the Gemini app — no app-building or per-integration approval required. By its Spring ’26 Edition on June 17, 2026, Shopify had removed the approval requirement for building on UCP entirely: developers self-serve by registering an agent profile and calling a public MCP endpoint.

The practical takeaway is unglamorous: publish a clean, structured, real-time product feed and let the agents read it, rather than hoping they scrape your pages correctly. That is where our eCommerce and headless-commerce team spends most of its agentic-readiness work — feed hygiene, live inventory and pricing, and protocol coverage — because it is the one input a merchant fully controls.

Almost every “checkout stays on-site” narrative quietly ignores Amazon, whose behavior is the opposite. On May 13, 2026 Amazon renamed Rufus — its assistant that reportedly helped 300M+ customers research, compare, and purchase in 2025 — to “Alexa for Shopping,” unifying it with Alexa+ in the main Amazon search bar. It shipped agentic purchase features: Auto-Buy (buy automatically when a tracked item hits a set price), Buy for Me (Amazon transacts on your behalf on other retailers’ sites for eligible products), Scheduled Actions, and up to a full year of price history. Amazon is doubling down on in-platform transactions, not backing away from them.

The tension is that Amazon is simultaneously fencing its own storefront off from competing agents — it sued Perplexity in November 2025 over its Comet browser scraping and buying on amazon.com, and has blocked “dozens” of agents from transacting on its site. Notably, Amazon endorsing the open UCP standard (for other merchants’ storefronts) in Shopify’s Spring ’26 backer list is not the same as Amazon opening amazon.com to third-party shopping agents. We unpack the merchant implications in what Amazon’s own shopping agent means for merchants.

None of this requires betting on which protocol wins. It requires preparing for the pattern that already emerged — strong AI discovery, weak in-chat checkout — and instrumenting for it. Four moves, in priority order.

If you want to pressure-test your own storefront against each of these, we keep a full agentic-commerce readiness checklist that turns the four moves into concrete, auditable line items — feed completeness, checkout ownership, AI-attribution tracking, and protocol coverage.