The in-chat checkout died. The discovery layer won.

In September 2025, OpenAI launched Instant Checkout: complete a purchase without leaving ChatGPT, starting with Etsy, then Shopify merchants, on a new Agentic Commerce Protocol built with Stripe. It was pitched as the future of shopping. By March 2026 it was gone — The Information broke the shutdown, OpenAI confirmed it, and reporting put the number of Shopify merchants that ever went live at fewer than fifteen. The company pivoted to what it calls a discovery-first experience: ChatGPT helps people find and compare products — by description, by uploaded image, by budget — and merchants keep the checkout.

Why the checkout failed

Buying inside a chat solved a problem few shoppers had. Checkout on a decent e-shop already takes a minute; what it offered in convenience it lost in trust, product detail, and the operational plumbing — returns, disputes, inventory accuracy — that real retail runs on. The transaction layer, it turns out, was already good enough. What was not solved — and what AI genuinely changes — is the step before: which product, which brand, why this one.

The discovery layer is where the value moved

The behavioural data points one way. According to Adobe’s March 2026 figures, AI-referred visitors to US retail sites now convert substantially better than other channels — a reversal from a year earlier — because they arrive having already compared, narrowed, and decided conversationally. The recommendation happens in the AI; the purchase happens with you. Whoever the assistant names wins the customer before the click exists.

An open flank most brands have not noticed

One structural detail deserves more attention than it gets: Amazon blocks OpenAI’s crawlers, so Amazon listings do not feed ChatGPT’s shopping answers. A brand’s own store can be present where the largest marketplace is absent. For smaller direct-to-consumer brands, that is a rare asymmetry — the giant chose not to play on a field that is growing.

What brands should take from six turbulent months

Platforms will keep redesigning the transaction — OpenAI pivoted once and will pivot again, Google is building its own agentic checkout. Chasing each mechanism is churn. The asset that survived every redesign is being the answer: clean, complete, machine-readable product data, and the third-party citations and content that make an assistant confident enough to recommend you. That work compounds regardless of whose checkout wins — and it is exactly the discipline we describe in GEO and test with the FIB AI Visibility Check.

How FIB approaches it

We treat AI commerce the way the evidence suggests: ignore the transactional plumbing until it settles, and invest in the discovery layer — the data, content, and authority that decide who gets named. If you want to know whether assistants recommend you or your competitors, get in touch.

Sources

CNBC, 24 March 2026 · The Information, 6 March 2026 (shutdown reporting) · OpenAI announcements, September 2025 – February 2026 · Adobe retail data, March 2026 · eMarketer on Google agentic checkout, November 2025 · Industry analyses (Elogic, Opascope, Ekamoira), 2026.