Walmart sold inside ChatGPT. Its checkout converted a third as well

Walmart put about 200,000 products on sale inside ChatGPT, saw a third of the conversion it gets by linking out, and took checkout back.

Walmart made about 200,000 products available for purchase directly inside OpenAI's ChatGPT from November 2025, and by March 2026 those in-chat sales were converting at one third the rate of the same shoppers clicking through to Walmart's own website, according to Walmart executive vice president Daniel Danker. Walmart then moved checkout back into its own shopping agent, Sparky, which now runs inside ChatGPT with the shopper's Walmart account and basket.

It is one of the few results a large retailer has put on the record about selling through a chatbot, and it matters now because ChatGPT is again being pitched to retailers, this time as an ad channel.

Why this is the question now

OpenAI began testing ads in ChatGPT in the US in February. On 31 August it opened self-service buying across 31 European markets, after the US and eight other markets, and it says tens of thousands of marketers have already advertised there. For the owner of a 20-store chain or a mid-sized online shop, the question has quietly changed from whether AI shopping is real to whether to spend money on it. Walmart is the company that already spent, with its own numbers attached.

What Walmart actually ran

Walmart announced the ChatGPT partnership in its newsroom on 14 October 2025. The version that went live in November was simple: a ChatGPT user asked for something, a Walmart product appeared in the answer, and the user could pay without leaving the chat. OpenAI took the shipping and payment details and placed the order. Larger items such as televisions were left out and still sent the shopper to Walmart's site.

What sold says a lot about who was shopping. Danker said vitamins and protein supplements were among the top sellers, and that more than half of in-chat orders came from automotive, beauty, home management, hardware and tools. Items that did well tended to be expensive enough to clear Walmart's shipping and small-basket fees on their own. These are the purchases people research first and buy one at a time, not the weekly shop.

Why the in-chat checkout lost

The problem was the basket. Instant Checkout completed an order for every single item. Walmart's shoppers, Danker said, worried that five separate purchases would arrive as five separate boxes, and did not want an order split away from the Walmart basket they already had going. The one-item design also broke something Walmart relies on at its own checkout: suggesting what goes with a purchase, such as cables with a television.

A chatbot can find the product. The retailer still has to own the basket, and that is where the sale is made or lost.

OpenAI said in March that the first version of Instant Checkout did not offer the flexibility it wanted, and that merchants could use their own checkouts while it concentrated on product discovery.

What Walmart changed

From the last week of March, a ChatGPT user who reaches Walmart meets Sparky, Walmart's own agent, running inside ChatGPT. The shopper signs in to their Walmart account once. The basket syncs across the Walmart app, the website and the chat, and payment uses the details Walmart already holds. Walmart said a similar setup would follow in Google's Gemini a month later.

Sparky is Walmart's own build. Danker described it as a mix of open-source models and retail models trained on Walmart's data, with questions routed to whichever model answers a given type best. He also said, unusually for an executive, that it is slow and gives weak answers often enough that some shoppers may write it off.

What did it produce?

Walmart has not published sales for the ChatGPT channel on its own, so the honest answer is partial. There are three things it has put on the record. Danker said ChatGPT sends Walmart about twice the rate of new customers that search engines do. On the 20 August 2026 earnings call, chief executive John Furner said the number of customers using Sparky was up 70% on a year earlier. He also said Sparky users spend 40% more per order than customers who do not use it. Earlier in the year Walmart had put that gap at about 35%.

Read the 40% with care. It compares people who choose to use Sparky with people who do not, and the people who plan a week of meals in a chat window were probably bigger spenders already. It shows Sparky users are valuable customers. It does not show Sparky made them spend more.

Should a 200-person retailer sell inside ChatGPT?

Not by building a checkout there. Walmart had a large catalogue, its own agent and a launch partnership with OpenAI, and it still found that in-chat checkout converted at a third of its usual rate. The part within your reach is being found: clean product data in a feed ChatGPT can read, with checkout left on your own site.

OpenAI's merchant documentation describes that feed: products are shared through an API or uploaded as a file to an SFTP server. That is catalogue work, much like a Google Shopping feed, not an AI project. Titles, attributes, stock and prices that are actually correct count for more than any chat feature.

What it does not mean

It does not mean AI shopping failed. Walmart kept the channel, kept the new-customer flow and changed who runs the checkout. It does not mean you need your own agent: Sparky exists because Walmart can spread the cost of building one across an entire national chain's shoppers, and a regional chain cannot. And the three-times figure is Walmart's own, from its own product mix. Your categories may do better or worse, and nothing yet published tells you which.

It also does not settle whether ChatGPT ads are worth buying. OpenAI now supports conversion tracking through its own pixel and a conversions API, so a test can at least be measured. But we could not find a retailer that has published its return on those ads, and Walmart's figures are about organic answers, not paid ones.

What to do this quarter

First, find out whether ChatGPT already sends you traffic: check your analytics for referrals from chatgpt.com and see what those visitors buy. Second, fix the product data before you spend anything, because that feed is also what an ad points to. Third, if you test ads, give the test one category that is researched before purchase and bought singly, like the ones that sold for Walmart, a fixed budget, and a decision date.

The verdict: Walmart's numbers make ChatGPT a place to be found, not yet a place to sell. Put the effort into your catalogue, keep checkout on your own site, and treat ads as a measured test, not a channel.