There is a special kind of arrogance in Silicon Valley that manifests as a belief that AI can improve literally everything. Search? AI will fix it. Writing? AI will fix it. Medical diagnosis? AI will fix it. And shopping, obviously, AI would absolutely revolutionize shopping. Just talk to your chatbot, tell it what you want, and it will handle the rest. Frictionless. Magical. The future.
Then Walmart looked at its actual conversion data, and the future collapsed.
Internal numbers revealed that OpenAI's Instant Checkout feature, the system that was supposed to let users buy products directly inside ChatGPT without ever leaving the conversation, was converting at roughly 3x lower than standard Walmart.com transactions. Three times lower. Not three percent lower. Three times. The gap between what OpenAI promised and what the data showed was so enormous that Walmart did what any rational business would do: it pulled back.
The Numbers That Killed Instant Checkout
Let's put this in perspective. E-commerce conversion rates are already brutally low by most standards. The average online retail conversion rate hovers around 2-3%. Walmart.com, thanks to years of optimization, a massive logistics network, customer trust built over decades, and a checkout flow refined through millions of A/B tests, performs well above that average. It took Walmart billions of dollars and years of engineering to build that checkout funnel.
OpenAI thought it could replicate that inside a chatbot. It could not. The Instant Checkout feature, which debuted with significant fanfare as part of OpenAI's push to make ChatGPT a commerce platform, produced conversion rates roughly three times lower than what Walmart sees on its own website. For a company that measures success in basis points of checkout improvement, this was not a disappointing result. It was a catastrophe.
The numbers made the business case impossible. Every shopper who started a purchase through ChatGPT's Instant Checkout instead of Walmart.com represented lost revenue. OpenAI was not adding a new channel; it was creating a worse version of an existing one. When you are Walmart, and your entire business runs on razor-thin margins and massive volume, routing customers through a checkout that converts at a third of your standard rate is the opposite of innovation. It is self-sabotage.
What Went Wrong: Why Nobody Buys Through a Chatbot
The failure is instructive because it exposes a fundamental misunderstanding about how people actually shop online. OpenAI's thesis was straightforward: people are already using ChatGPT for product research, so why not let them buy right there in the conversation? Reduce the steps, reduce the friction, increase the conversions. Classic Silicon Valley logic. It sounds great in a pitch deck.
In reality, the checkout experience inside ChatGPT introduced more friction, not less. Users had to trust an AI intermediary with their payment information. They could not see the familiar Walmart interface they were accustomed to. Product images and details were stripped down to fit a conversational format. The shopping cart lived inside a chat window, not the purpose-built cart system that Walmart has spent years perfecting. Return policies, shipping options, payment methods, all of it felt unfamiliar and disconnected from the Walmart experience users already trusted.
There is also the matter of intent. When someone goes to Walmart.com, they are in shopping mode. They have a credit card nearby. They are mentally prepared to buy. When someone is chatting with ChatGPT, they are in conversation mode. They might ask about a product, but the mental shift from "I'm having a chat" to "I'm going to enter my payment details" is more jarring than OpenAI anticipated. The conversion funnel psychology is completely different, and OpenAI's Instant Checkout ignored all of it.
The data told the story clearly. Shoppers would browse, maybe add items to a cart, and then abandon the process at vastly higher rates than on Walmart.com. The AI middleman was not removing friction. It was the friction.
OpenAI's Retreat: Instant Checkout Is Dead
Faced with the Walmart pullback and presumably similar data from other retail partners, OpenAI is making a significant strategic retreat. The company is ending Instant Checkout entirely. The feature that was supposed to turn ChatGPT into a commerce juggernaut is being replaced by a different model: retailer-dedicated apps inside ChatGPT.
This is a revealing pivot. OpenAI originally positioned Instant Checkout as a way to own the entire shopping experience. The user would never leave ChatGPT. OpenAI would sit between the customer and the retailer, collecting data on purchasing behavior, building an advertising business, extracting value from every transaction. It was an enormously ambitious play for a company that, until recently, was known primarily for text generation.
The new approach is the opposite. Instead of OpenAI controlling the checkout, retailers will build their own apps that live inside ChatGPT but operate independently. Users will interact with the retailer's AI, log into the retailer's account, and complete the purchase through the retailer's systems. OpenAI becomes the platform, not the middleman.
OpenAI's retreat from Instant Checkout is a tacit admission that the company overestimated how much consumers would trust an AI chatbot with their money. The data did not lie: shoppers overwhelmingly preferred the checkout systems they already knew.
It is the kind of strategic reversal that gets described as an "evolution" in press releases but looks a lot more like "we tried to control the checkout and the numbers were terrible so now we're giving that control back to the people who actually know how to sell things."
Sparky Enters the Chat: Walmart Takes Control
Walmart is not abandoning the idea of AI-assisted shopping through ChatGPT. It is abandoning the idea of OpenAI handling the shopping part. Starting the week of March 25, 2026, Walmart is integrating its own Sparky AI assistant into ChatGPT. The difference is architectural and it matters enormously.
Under the old model, OpenAI's system processed the transaction. Under the new model, users interact with Sparky inside ChatGPT but log into their Walmart accounts directly. Walmart keeps control of the underlying checkout systems. The payment processing, the order management, the shipping logistics, the customer service, all of it stays on Walmart's infrastructure. ChatGPT becomes the conversation layer, not the commerce layer.
This might sound like a subtle distinction, but it changes everything. Shopping carts are now synchronized across Walmart.com, the Walmart mobile app, and the ChatGPT Sparky experience. If you add something to your cart through Sparky, it appears in your regular Walmart cart. If you decide to complete the purchase on Walmart.com instead, the items are waiting for you. The experience is integrated with Walmart's existing ecosystem rather than siloed inside ChatGPT's.
Walmart, to its credit, recognized something that OpenAI did not: the value of a shopping AI is in the discovery and recommendation phase, not in replacing a checkout that already works. Sparky can help you find products, compare options, make recommendations based on your purchase history. But when it is time to actually buy, you are dealing with Walmart's trusted checkout, not an AI experiment.
Early Results: Better, But Still Not Close
The early numbers for Sparky inside ChatGPT are better than Instant Checkout, but they also tell a sobering story. Sparky is converting at roughly 70% of the Walmart.com rate. That is a massive improvement over the three-times-lower performance of Instant Checkout, but it still means that for every 10 purchases that would happen on Walmart.com, only 7 happen through the Sparky-in-ChatGPT experience.
A 30% gap is still significant. It suggests that even with Walmart controlling the checkout and the cart synchronization and the familiar account system, there is something about the chatbot context that reduces purchasing intent. Maybe it is habit. Maybe people who want to buy something still prefer to open Walmart.com or the app directly. Maybe the conversational interface, no matter how sophisticated, will always lag behind a purpose-built shopping experience.
Or maybe 70% is just where it starts, and the number will climb as users get comfortable with the integrated experience. That is certainly the bet Walmart is making. But it is worth noting that even the improved model, the one where Walmart controls everything, still cannot match what Walmart already has without any AI involvement at all.
The Bigger Problem With AI Commerce
The Walmart-OpenAI checkout debacle is not just a story about one failed feature. It is a case study in the gap between AI hype and commercial reality. OpenAI has been aggressively positioning ChatGPT as a platform for everything: search, productivity, creative work, customer service, and now shopping. The company's valuation depends on ChatGPT becoming indispensable across every domain of digital life.
But e-commerce is not like search or writing assistance. When you ask ChatGPT a question and the answer is mediocre, you shrug and try a different prompt. When you are about to spend $200 on groceries and the checkout experience feels unfamiliar or unreliable, you close the tab and go to the website you trust. The stakes are higher. The tolerance for friction is lower. The switching cost is nothing, because every retailer has its own website that works perfectly fine.
The Instant Checkout Timeline
This is the recurring theme of 2026 in AI: companies discovering that the real world is harder than the demo. ChatGPT can generate a convincing product description, but it cannot convince a shopper to trust it with a credit card number. It can recommend the perfect pair of running shoes based on your preferences, but it cannot replicate the confidence a user feels when they see the familiar Walmart checkout page with its secure payment badges and the free shipping threshold they know by heart.
OpenAI's pivot to retailer-dedicated apps is an acknowledgment of this reality, even if the company would never frame it that way publicly. The lesson is clear: AI can enhance the shopping experience at the discovery and decision-making stage, but it is not ready to replace the infrastructure that makes people comfortable spending money online. That infrastructure was built over 20 years by companies that understand consumer trust at a granular level. OpenAI thought it could skip all of that. It could not.
The question now is whether OpenAI's platform play, hosting retailer apps inside ChatGPT, will generate enough value to justify the company's ambitions in commerce. Walmart, for its part, seems willing to experiment. But the power dynamic has shifted. Walmart is no longer a partner handing its checkout to OpenAI. It is a tenant running its own store inside OpenAI's mall, and it brought its own cash registers.
What Actually Works: AI Shopping Done Right
Here is the thing that makes OpenAI's Instant Checkout failure even more embarrassing: other AI platforms figured out how to do commerce without trying to be the middleman. While OpenAI was inserting itself between the customer and the retailer, creating unnecessary friction and tanking conversion rates, competitors were building smarter integrations that actually respect how people shop.
Google's Gemini shopping integration is probably the clearest contrast. Instead of trying to own the checkout experience, Gemini plugs directly into Google Shopping's existing infrastructure. Users search for products through Gemini, see real-time pricing and availability, and when they are ready to buy, they are routed through Google Shopping's native checkout, a system that already has their payment information, shipping addresses, and purchase protection built in. There is no AI middleman awkwardly trying to process a credit card. The AI handles the discovery. The established commerce system handles the transaction. It is not revolutionary. It is just sensible.
Then there is Perplexity, which launched its own shopping feature that takes a completely different approach. Perplexity lets users buy products directly with one click inside the platform. The key difference is that Perplexity built its buying experience to be absurdly simple, stripping away every unnecessary step rather than adding new ones. Users have praised it for being the most frictionless AI shopping experience available, precisely because it does not try to replicate an entire retail checkout flow inside a chat window. It just gets out of the way.
The pattern is clear: AI platforms that succeed at commerce are the ones that do not try to replace existing checkout systems. Google Gemini leans on Google Shopping. Perplexity built a streamlined one-click experience. OpenAI tried to be the checkout itself, and the data showed exactly how badly that backfired.
What all of this reveals is that OpenAI's fundamental error was not a technical one. It was a strategic one. The company looked at commerce and saw an opportunity to own the entire transaction, to sit between every buyer and every seller, extracting data and value from every purchase. Google and Perplexity looked at the same opportunity and realized that AI's value in shopping is in helping you find the right product, not in handling your money. Discovery is where AI shines. Checkout is where trust matters. And trust is something you earn over years, not something you bolt onto a chatbot.
If you have tried shopping through ChatGPT and found the experience frustrating, clunky, or just vaguely unsettling, you are not wrong. The data validates your instinct. But the failure of ChatGPT shopping does not mean AI-assisted commerce is dead. It means OpenAI chose the wrong approach. Other platforms are doing it better right now, and the gap is only going to widen as they refine their integrations while OpenAI scrambles to figure out what its retailer app platform even is.
The Verdict
OpenAI's Instant Checkout failed because it tried to replace trusted retail checkout systems with an AI intermediary that shoppers did not trust. Walmart's pivot to Sparky is an improvement, but the 30% conversion gap shows that even the best AI shopping integration still cannot match what retailers built without AI at all. The future of AI commerce is not replacing checkout. It is admitting that checkout was never the problem.
Somewhere in Bentonville, Arkansas, a Walmart executive is looking at the Sparky conversion data and probably thinking what every retailer has thought since OpenAI started its commerce push: "We could have told you this would happen." The companies that spent decades building trusted online shopping experiences understood something that OpenAI, in its rush to become everything to everyone, completely missed. People do not want a smarter checkout. They want the checkout they already know. And no amount of AI magic is going to change that overnight.