Walmart is putting its Sparky chatbot inside ChatGPT and Gemini. Not as a plugin. Not through OpenAI's commerce infrastructure. As its own AI, embedded in someone else's platform. The reason is in the same headline: OpenAI's Instant Checkout — the feature that was supposed to let users buy without leaving ChatGPT — had conversion rates three times lower than simply clicking out to the retailer's site. Three years after ChatGPT plugins launched with a vision of the AI interface as storefront, the checkout went back to the store.
The Storefront
OpenAI has been building toward commerce inside ChatGPT since the beginning. The arc is clear in retrospect.
-
MAR 2023ChatGPT plugins launch. Expedia, Instacart, Kayak, Klarna, OpenTable, Shopify, Wolfram. Stratechery calls it "ChatGPT as aggregator" — a major shift in ambition.
- Apr 2025 OpenAI rolls out product recommendations in ChatGPT with buy buttons — but the buttons link out to merchant websites.
-
JUL 2025The Financial Times reports OpenAI aims to integrate a checkout system into ChatGPT "to ensure it captures a cut of purchase transactions." The goal is explicit: own the transaction, take a percentage.
- Sep 2025 OpenAI launches Instant Checkout — single-item purchases without leaving ChatGPT.
- Oct 2025 OpenAI announces apps inside ChatGPT: Booking.com, Canva, Expedia, Spotify, Zillow.
- Dec 2025 OpenAI and Instacart launch grocery shopping inside ChatGPT.
Each step moved in one direction: from linking out (buy buttons) to keeping users in (Instant Checkout). The ambition was to make ChatGPT the place where discovery AND transaction happen. An AI storefront where users browse, choose, and pay — all inside the chat interface. OpenAI would take a cut of every purchase, the way Apple takes a cut of every app sale.
Three Times
It didn't work.
By January 2026, The Information reported that OpenAI had been slow to expand in-app checkouts. On March 5, The Information reported that OpenAI was scaling back shopping directly inside ChatGPT, routing checkouts back to the apps that plug into the platform. Two weeks later, the Walmart headline confirmed why: users who discovered products through ChatGPT converted at one-third the rate when checking out inside the AI versus clicking through to the retailer's own environment.
The gap isn't surprising once you name it. Discovery and transaction have different trust requirements. A user will happily ask ChatGPT "what's a good air fryer under $80?" But typing a credit card number into a chat interface — a format designed for conversation, not commerce — is a different act. The storefront needs to feel like a store. ChatGPT feels like a conversation.
The Reversal
Walmart's response is the inversion of everything OpenAI was building toward. Instead of Walmart being a merchant inside OpenAI's commerce platform, Walmart is bringing its own AI into OpenAI's audience. Sparky — Walmart's chatbot — will live inside ChatGPT and Gemini, handling the shopping conversation on Walmart's terms, presumably routing transactions through Walmart's own checkout.
The distinction matters. When OpenAI owned the checkout, OpenAI controlled the transaction — the price display, the payment flow, the data, the cut. When Walmart brings Sparky, Walmart controls all of that. ChatGPT becomes the place where users encounter Sparky, not the place where they buy from Walmart.
In platform terms: ChatGPT tried to be a storefront and became a mall. The difference is who owns the register. A storefront takes a cut of every sale. A mall provides foot traffic and charges rent. OpenAI's Instant Checkout was the storefront model. Walmart embedding Sparky is the mall model. The retailer moved its own shop inside the building and kept the keys to the cash register.
And Walmart is going multi-platform — Sparky in ChatGPT and Gemini both. The same way a retailer opens locations in multiple malls. The AI platforms compete for tenants.
The Landlord
One more number from March 19, in a Wall Street Journal piece that arrived alongside the Walmart story.
Apple's GenAI app revenue nearly tripled in seven months, with three-quarters coming from a single app: ChatGPT. Apple is on pace to surpass $1 billion in AI revenue this year. Not from building AI. Not from training models. Not from operating data centers. From its 30% cut of App Store subscriptions — most of which are ChatGPT Plus and Pro users paying through iOS.
The irony is structural. OpenAI's July 2025 ambition was to "capture a cut of purchase transactions" — to be the entity that taxes commerce flowing through AI. But that's Apple's business model, and Apple has been executing it since before ChatGPT existed. OpenAI wanted to be the platform that takes a percentage. Apple already is.
As ChatGPT becomes the mall — with Walmart, Instacart, Booking.com, and others bringing their own shops inside — usage grows. Subscriptions grow. And Apple's cut of those subscriptions grows with it. The more valuable ChatGPT becomes as a commerce platform, the more Apple earns. The landlord doesn't need to own the stores or the checkout. The landlord owns the building.
The Register
Three years from plugins to this. OpenAI launched ChatGPT plugins in March 2023 with an aggregator's ambition — every retailer, every booking engine, every service accessible through a single conversational interface. The vision was that AI would replace the web browser as the gateway to commerce. It was right about that. What it got wrong was who would own the register.
The answer, as of March 19, 2026: not the AI. Users discover through ChatGPT. They buy through Walmart. And Apple collects rent on the way in.