Since November, Walmart has let some ChatGPT users order a limited selection of products without ever leaving OpenAI’s chatbot interface. Sales have been disappointing, a Walmart executive vice president exclusively tells WIRED.
The results suggest that a future where chatbots and AI agents take over ecommerce is still a way off, if it ever materializes. Last year, OpenAI made a bet that it could boost revenue by charging a commission on purchases made through ChatGPT. It partnered with Walmart, Etsy, and other shops on an “agentic commerce” feature called Instant Checkout.
Walmart has made about 200,000 products available directly in chat responses, allowing consumers to provide their shipping and payment details to OpenAI and place their order within ChatGPT. For products like TVs, shoppers still have to open Walmart’s website to make a purchase the old-fashioned online way. Conversion rates—the percentage of users following through with a purchase of an item shown to them by ChatGPT—have been three times lower for the selection sold directly inside the chatbot than those that require clicking out, according to Daniel Danker, who oversees design and product for Walmart. Put simply, Instant Checkout has been a flop.
OpenAI and Walmart could have spent years trying to fix the ”unsatisfying” consumer experience of Instant Checkout, Danker says. He credits OpenAI for choosing instead to quickly move to a new system long favored by Walmart. Next week, Walmart’s chatbot, Sparky, will begin operating within ChatGPT—essentially a chatbot inside a chatbot. A similar setup will arrive in Google’s Gemini chatbot next month.
The approach solves what Danker says he believes is the biggest problem with Instant Checkout: It forces people to buy items individually. “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 says. “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.”
Among the items that have been available in the Instant Checkout experience, top sellers include vitamin and protein supplements. People new to GLP-1 weight-loss drugs asking ChatGPT what they need to know about them are receiving advice to increase nutrient intake, Danker says. Other items that have done well tended to be pricey enough that the orders weren’t subject to added shipping or small-basket fees. All told, the automotive, beauty, home management, hardware, and tools categories account for over half of Instant Checkout orders.
In the new experience, Walmart users log into Sparky the first time they encounter it in ChatGPT. Their basket from Walmart’s website or app and within ChatGPT will sync with another in the hopes of better reflecting people’s actual shopping habits. Consumers add peanut butter one day on the Walmart app, foil the next, and a birthday gift at the last second on the website before checking out. “When Sparky travels, it’s the Walmart store meeting you where you are, instead of a completely broken experience,” Danker says.
Walmart has good reason to want to get the experience in ChatGPT correct. The chatbot is now bringing in about twice the rate of new customers as search engines, Danker says. He suspects that’s because the power users of ChatGPT are not typical Walmart customers. But the retailer’s price, selection, and broad geographic footprint mean that its products are showing up in many ChatGPT responses.
Sparky was developed by Walmart, Danker says. But it relies on open source generative AI models combined with some retail-specific ones 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 says. “It’s never stuck in any one.”