OpenAI Tried to Own the Cart. Walmart Took It Back.

In October 2025, Walmart and OpenAI announced what both companies described as a major step in the future of e-commerce. Customers would be able to shop Walmart directly inside ChatGPT using a feature called Instant Checkout.

No browser tab, redirections, or even the need to leave the conversation. OpenAI CEO Sam Altman and Walmart CEO Doug McMillon were both publicly enthusiastic about the partnership.

Five months later, Walmart is pulling the plug.

Conversion rates for products sold directly inside ChatGPT were three times lower than those for products that required users to click out to Walmart's own website, according to Daniel Danker, Executive Vice President of AI Acceleration, Product, and Design at Walmart.

The data, disclosed to WIRED in its March 18th report, represents one of the first hard performance disclosures from a major retailer testing AI-mediated commerce.

OpenAI has confirmed it is ending Instant Checkout. Starting the week of March 25, Walmart's own chatbot, Sparky, will begin operating inside ChatGPT. A similar integration will arrive in Google's Gemini the following month. The shift hands control of the shopping experience back to Walmart, and away from OpenAI.

Why Instant Checkout Failed

The core problem with Instant Checkout was structural. OpenAI obtained product data by scraping partner retailers' websites, which meant it could not verify whether products were in stock or confirm accurate delivery times. The checkout process forced users to buy items individually, with no ability to sync with an existing Walmart cart.

"They fear that when checkout happens automatically after every single item that they're going to receive five boxes when they actually just want it all in one," Danker said. "They generally don't want to split the checkout experience, where it buys the one item, even though they had other items in their Walmart cart already."

Emily Pfeiffer, Principal Analyst at Forrester, told CNBC that the approach was insufficient from the start. "Crawling and scraping is inadequate to get the full breadth of product data that you need to do a good job of commerce," Pfeiffer said.

Bob Hetu, analyst at Gartner, said the failure reflected a broader miscalculation. "OpenAI underestimated how difficult the enablement of transactions was going to be, which, on the one hand, is a little surprising, but on the other hand, it's not easy for retailers," Hetu told CNBC.

The categories that performed best in Instant Checkout like vitamin and protein supplements, automotive, beauty, home management, hardware, and tools were products pricey enough to avoid added shipping or small-basket fees. Even within those categories, conversion remained well below what Walmart sees on its own platform.

Walmart's fix is to replace OpenAI's checkout infrastructure with its own. Sparky, Walmart's in-house shopping assistant, has been live on Walmart's website and mobile apps since 2025. It runs on a combination of open source generative AI models and retail-specific models trained on decades of Walmart data.

"We're able to route certain questions to one model and certain questions to another because we find that the quality of answers differs," Danker said to WIRED. "It's never stuck in any one."

In the new setup, Walmart users log into Sparky the first time they encounter it in ChatGPT. Their basket from Walmart's website, the Walmart app, and ChatGPT will synchronize, allowing purchases to reflect ongoing shopping behavior rather than a single isolated transaction.

"When Sparky travels, it's the Walmart store meeting you where you are, instead of a completely broken experience," Danker said.

Early conversion data from the new setup is more encouraging. Users who access Sparky through ChatGPT complete purchases at approximately 70% of the rate of those who shop directly on Walmart.com, according to TheStreet. That figure, while still below Walmart's own platform, is substantially higher than what Instant Checkout produced.

Sparky also carries behavioral data. Danker said ChatGPT is now bringing in roughly twice the rate of new Walmart customers as search engines. He attributed this to the profile of ChatGPT's power users, who skew toward demographics not typically associated with Walmart's core customer base. The retailer's price, selection, and geographic footprint mean its products appear frequently in ChatGPT responses regardless.

Half of Walmart's app users have engaged with Sparky, and Walmart US CEO David Guggina said Sparky users spend approximately 35% more per order than other shoppers.

Own the Agent, Rent the Distribution

The shift in the Walmart-OpenAI relationship shows a pattern emerging across retail. Rather than allowing AI platforms to own the checkout experience, retailers are embedding their own branded agents inside those platforms, keeping customer data, the transaction, and the post-purchase relationship in-house.

OpenAI gets the distribution. The retailer keeps the customer.

Walmart is not alone in this direction. Target and Instacart have also moved away from Instant Checkout in favor of their own embedded retail apps inside ChatGPT. OpenAI spokesperson Taya Christianson confirmed the company is now focused on improvements that give merchants more control over checkout.

"We appreciate our partners for learning with us," Christianson said.

The pattern also shows a data based decision. Instant Checkout gave retailers visibility into a transaction only after a purchase was made. An embedded app gives them visibility into the entire discovery and decision-making process, from the first query to checkout.

While Walmart embeds Sparky in ChatGPT and partners with Google's Gemini, Amazon has taken the opposite approach. The company recently won a temporary court order barring Perplexity's automated technology from accessing its site and has shown no indication it will integrate with ChatGPT or other AI assistants.

Amazon's absence is strategic. With Alexa, 40% of US e-commerce market share, and control over product discovery through Amazon Search, the company has no incentive to route customers through external AI platforms.

When you dominate e-commerce, you can afford to block AI agents. When you're competing to close that gap, you cannot. Walmart has explicitly said it will not block AI agents from shopping on its platform.

The Limits of Agentic Commerce

Walmart's data is consistent with what broader consumer research shows. A survey by Adobe-owned Semrush found that only 22% of US consumers have bought a product inside an AI tool. Half said they made a purchase after using AI during research.

The data shows a consistent pattern. AI is functioning effectively as a discovery channel, helping consumers find products, compare options, and form intent. It is not yet functioning effectively as a transaction channel. Consumers are willing to research inside a chatbot but they are not yet willing to hand over the checkout.

Danker acknowledged the limits of the vision that Instant Checkout was built on. "This idea that it will all become automated might be a little bit far-fetched," he said to WIRED. "People do get excited about shopping for clothes, for their home, for their children."

The Sparky integration into ChatGPT and Gemini does not resolve that underlying problem. It repositions Walmart to benefit from AI as a discovery channel while keeping checkout inside infrastructure it controls.

What Sparky's expansion does establish is that the first major model for agentic retail commerce has failed its first significant test. Under the new model, AI platforms handle discovery and retailers handle the transaction.