What Walmart's ChatGPT Pivot Proves About AI Commerce

Walmart tried letting OpenAI own the checkout. It failed. What happened next is a blueprint for the entire industry.

In October 2025, Walmart announced a partnership with OpenAI to let shoppers buy products directly through ChatGPT's Instant Checkout feature. It was billed as the future of shopping — conversational commerce where you never leave the chat.

A few months later, Walmart pulled the plug.

The feature struggled with accuracy. Wrong items ended up in carts. Conversion rates were significantly lower than Walmart's own channels. The promise of frictionless AI-powered checkout ran headfirst into a fundamental reality: the AI platform wasn't good enough at the last mile to justify giving up control.

But Walmart didn't retreat from AI commerce. They pivoted to something far more interesting — and far more instructive for anyone thinking about the future of product discovery.

The pivot: own the agent, rent the distribution

Instead of letting OpenAI own the shopping experience, Walmart embedded its own AI assistant, Sparky, directly into ChatGPT and Google Gemini. Same distribution channels (hundreds of millions of users), fundamentally different architecture.

Now when a ChatGPT user asks about Walmart products, they interact with Walmart's agent inside OpenAI's interface. Walmart controls the customer data, the product recommendations, the shopping experience, and the transaction. OpenAI provides the eyeballs.

Early results from Walmart's pilot reportedly show users completing purchases at roughly 70% of the rate of those using Walmart.com directly — far better than the Instant Checkout numbers. The key difference appears to be trust: customers recognize they're interacting with Walmart's agent, even inside another app.

This is a pivotal moment. The world's largest retailer just demonstrated, through expensive trial and error, that the right architecture for AI commerce is not "AI does everything." It's "AI provides the reach, the brand provides the intelligence."

The data behind the shift

The numbers tell the story of why this matters:

ChatGPT now drives roughly 20% of Walmart's referral clicks, up significantly from just months prior. Other retailers are seeing similar trends — Etsy above 20%, Target near 15%, eBay around 10%. AI tools are becoming a genuine discovery channel.

Meanwhile, Amazon has conspicuously stayed out. They've actively blocked AI crawlers from their product data. They have their own AI assistant, Rufus, and a $60 billion advertising business built on owning discovery. Amazon has no reason to let anyone else into that funnel.

This creates a fascinating asymmetry. Retailers who aren't Amazon are desperate for new customer acquisition channels. AI platforms are desperate for commerce revenue. But the first attempt at connecting them — give the AI platform the checkout — failed.

What this proves about discovery vs. conversion

Walmart's pivot reveals a clean line between two fundamentally different capabilities:

Discovery — understanding what a customer wants, surfacing relevant options, and narrowing the field. This is where AI excels. Natural language understanding, visual similarity, preference learning, and constraint handling are all tasks where machine learning outperforms keyword search by orders of magnitude.

Conversion — the final recommendation, the trust signal, the checkout experience, the post-purchase relationship. This is where brands excel. Customers trust Walmart to get their order right. They trust their favorite fashion brand to understand quality and fit. They trust their insurance agent to explain the fine print.

The Instant Checkout failure happened precisely at this boundary. OpenAI tried to own both discovery and conversion. Customers got wrong items in their carts because the AI didn't have the deep product intelligence that Walmart's own systems have. Conversion suffered because the trust wasn't there.

The Sparky pivot succeeds because it respects the boundary. AI handles the top of funnel (where ChatGPT's massive user base provides reach), and Walmart handles everything from product intelligence through checkout (where its systems and brand trust provide reliability).

The context handover model

Walmart's pivot is a specific instance of a broader architectural pattern that we believe will define the next generation of AI commerce: the context handover.

Here's how it works:

AI handles discovery. A customer interacts with an AI that understands their intent, refines their preferences through conversation, filters options based on constraints, and learns their taste — all in about 60 seconds. No login. No cookies. No purchase history required.
Context is structured and passed forward. Everything the AI learned — what the customer wants, what they've considered, what they've rejected, and why — is packaged into a structured context object.
The brand receives a qualified customer. Instead of a cold click from a search ad, the brand gets a customer whose needs have already been understood. The journey doesn't reset. The brand can personalize from the first moment, because the hard work of understanding has already been done.

This is fundamentally different from both the old model (keyword search → product grid → hope they buy) and the agentic model (AI does everything, brand is just a fulfillment endpoint).

In the context handover model:

The AI earns as a referrer — compensated for the discovery value it creates
The brand keeps the customer relationship, the data, and the transaction
The customer gets a dramatically better experience because nothing resets between discovery and purchase

Walmart arrived at a version of this through failure. The question is whether the rest of the industry can skip the "let the AI platform own checkout and watch it fail" phase and go straight to the architecture that works.

Implications for every retailer and brand

If you're a retailer or brand watching the Walmart saga, here are the takeaways:

AI is a discovery channel, not a sales channel. The data is clear — consumers are using AI for product research and inspiration. But they still want human-controlled, brand-controlled experiences for the actual purchase. Build for that reality.

Own your AI intelligence. Walmart's pivot was about taking back the agent. If your product discovery experience is going to live inside ChatGPT, Gemini, or any other AI platform, the intelligence powering it should be yours — tuned to your catalog, your customers, and your brand.

The referral window won't stay free. Independent analysts have been clear: the current wave of free AI referral traffic is temporary. AI platforms will monetize. The brands that build their own AI discovery capabilities now will be less dependent on whatever toll system OpenAI or Google eventually impose.

Context is the new cookie. In a world where third-party cookies are dying and privacy regulations are tightening, structured context from an AI discovery session is an incredibly valuable signal. A customer who arrives with "I want a mid-century modern dining table, under $2,000, seats 6, in walnut, and I've already ruled out West Elm" is worth ten anonymous pageviews.

The future of AI commerce isn't about who owns the checkout button. It's about who builds the best handover between AI-powered discovery and brand-owned conversion.

Walmart just spent millions learning that lesson. The rest of the industry should take notes.

This post references reporting from Wired, Modern Retail, CNBC, Retail Dive, and analysis from Juozas Kaziukėnas and Eric Seufert.