You ask ChatGPT to help you pick a laptop. It compares specs, reads reviews, narrows your list to three options in under a minute. Genuinely useful. The kind of moment where you think: maybe I never need to open a browser tab again.

OpenAI thought the same thing — except they wanted to own what happens after you pick the laptop too. The logic is simple math. OpenAI burns billions per year. Subscriptions alone don't justify a $300B valuation. They need a transaction layer — a way to turn conversations into revenue. Not just answering questions. Closing sales.

The Checkout Experiment

So OpenAI shipped Instant Checkout and the Agentic Commerce Protocol (ACP) — a standard that lets AI assistants handle purchases on behalf of users, like a digital shopping concierge. Walmart, Target, Sephora, Best Buy, Home Depot, Nordstrom, Wayfair — all signed on starting November 2025. The pitch: users buy products without leaving the chat window. One conversation, one purchase, zero friction.

In practice, a mess. Walmart offered roughly 200,000 products through Instant Checkout. The system scraped retailer websites to surface items but couldn't verify real-time stock levels or delivery times. It forced single-item checkout — meaning if you wanted shampoo, toothpaste, and paper towels, you'd place three separate orders and receive three separate boxes. Daniel Danker, Walmart's EVP of AI Acceleration, told WIRED the experience was "unsatisfying." Customers feared "five boxes when they actually just want it all in one." No coupons, no loyalty points, no store pickup. The basics of e-commerce, missing entirely.

The numbers confirmed the gut feeling. In-chat purchases converted at one-third the rate of click-out transactions to Walmart.com. A 3x conversion gap. Etsy, another launch partner, reported it didn't achieve "a large volume of sales" either. Users happily researched products inside ChatGPT. They just refused to buy there.

The Retreat

Walmart fired OpenAI's checkout entirely and plugged in Sparky — Walmart's own AI chatbot — directly into ChatGPT instead. Sparky handles the full Walmart experience: cart building, coupons, loyalty, delivery scheduling. Walmart keeps the customer data, the transaction, and the post-purchase relationship. OpenAI gets a flat API fee. No revenue share. No margin.

On March 24, OpenAI made it official: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide." Corporate translation: it flopped. They're pivoting ACP to discovery-only — meaning ChatGPT can help you find products, but someone else handles the actual money part.

Meanwhile, Google's Universal Commerce Protocol — announced at NRF in January 2026 — supports full embedded checkout from day one, with Shopify, Stripe, Visa, and Mastercard already on board. Google watched OpenAI fail, took notes, and shipped the version that doesn't suck.

The Ceiling

The tradeoff is now obvious. OpenAI controls product discovery for major retailers but owns zero transaction data, zero margin, zero checkout. Every retailer keeps their funnel, their loyalty program, their payment rails. ChatGPT brings Walmart roughly 2x the rate of new customers compared to search engines — real discovery value. But discovery without checkout is just advertising. Expensive, AI-powered advertising.

If you're building on ChatGPT's commerce layer, you're building on a referral engine. The moment every retailer embeds their own AI assistant on every platform, the middleman value evaporates. Walmart already proved it — they didn't need OpenAI's checkout, they needed OpenAI's audience.

ChatGPT became Google Shopping with better conversation skills. For a company that needs to justify a $300B valuation, that ceiling is uncomfortably low.