Launched with fanfare in autumn 2025 as "the next step for agentic commerce," Instant Checkout, the feature that allowed users to buy products directly from ChatGPT, was quietly buried just months after its launch. OpenAI is pivoting, its retail partners are adapting, and the entire industry is drawing lessons from an experiment that exposed the real difficulties of shopping via artificial intelligence.
Key takeaways:
- Instant Checkout is officially over. OpenAI abandons its native payment system in ChatGPT in favor of dedicated apps that redirect the user to the retailer's site to complete the purchase.
- The reasons for the failure are multiple: limited catalog, product data often outdated, frequent errors and very low conversion rates — three times lower than purchases made directly on a retailer's site, according to Walmart.
- Retailers are taking back control. Walmart integrates its own assistant, Sparky, into ChatGPT and Gemini; Etsy is developing its own ChatGPT app. The trend is clear: major brands want to control the customer experience and the data, not delegate them to OpenAI.
- No one has found the formula yet. According to analysts, while shoppers increasingly use chatbots to research products, they do not yet complete their purchases within the conversational interface.
An ambition proclaimed with great fanfare, a much more complex reality
When OpenAI unveiled Instant Checkout in autumn 2025, the world of online commerce held its breath. The idea: let ChatGPT users buy products without leaving the chat window. Etsy, Walmart and Shopify rushed to join the effort. Harley Finkelstein, president of Shopify, called it a "new frontier" for online retail.
Five months later, the tone changed radically. Daniel Danker, executive vice president of AI at Walmart, summed up the situation with disarming frankness at Morgan Stanley’s Tech, Media & Telecom conference in early March 2026: Instant Checkout, he said, was "just a very temporary moment." He added that in less than a month, this experience would be gone.
Why didn't it work?
Instant Checkout's problems are both technical and structural. To obtain product data, OpenAI relied in part on scraping and crawling retailers' sites. The result: stock information, delivery times or prices were frequently inaccurate or out of dateEmily Pfeiffer, senior analyst at Forrester, is blunt: " Crawling and scraping are not enough to obtain the depth of product data necessary for a satisfactory shopping experience. "
The integration of merchants also proved far more laborious than expected. Despite initial announcements promising more than a million Shopify merchants, only around 30 were available via Instant Checkout at the time the feature was abandoned. Walmart had put about 200,000 products online; a significant number, but insufficient to cover the reality of a modern retail catalog.
Bob Hetu, an analyst at Gartner explains bluntly: " OpenAI underestimated the difficulty of enabling transactions. Which, on the one hand, is a bit surprising, but on the other, it’s not simple for retailers either. "
The killer data: conversion rates
The figures shared by Walmart are particularly telling. Conversion rates for products sold directly in ChatGPT via Instant Checkout were three times lower to those recorded when the user was redirected to the retailer’s site to complete their purchase. An Adobe-Semrush study published in March 2026, based on a panel of more than 1,000 American consumers, confirms the trend: only 22% of users have already purchased a product directly within an AI tool. Conversely, half say they made a purchase after using AI to research.
The message is clear: chatbots have become powerful tools for product discovery and research, but not yet for completing purchases.
The new model: retailers keep the keys
Facing this failure, OpenAI and its partners chose a radically different approach. Rather than centralizing the transaction in ChatGPT, the new model relies on dedicated applications, so-called "ChatGPT Apps", developed by each retailer. When a user finds a product via the chatbot, they are redirected to the brand’s site to complete their purchase, either in a browser embedded in the ChatGPT mobile app or in a separate web tab.
Walmart best illustrates this pivot. The retail chain integrates its own shopping assistant, Sparky, directly into ChatGPT and Google Gemini. The mechanics are simple but strategic: when a user asks ChatGPT to find a deal on a product, the request is routed to Sparky, which queries Walmart’s inventory, presents options and processes the order, all while remaining inside the ChatGPT interface. OpenAI gains exposure and potentially access fees, but Walmart retains the customer data, the transaction and the post-purchase relationship.
The strategic stake is significant: " We have learned that our customers want consistency at every touchpoint," a Walmart spokesperson said. Data from Sparky pilot tests in ChatGPT show that users who access it complete their purchases at roughly 70% of the rate observed directly on Walmart.com, a much more satisfactory result than Instant Checkout. The key, according to Walmart, lies in trust: customers know they are interacting with Walmart's agent, even within another app.
Etsy adopts the same logic. The marketplace is developing its own ChatGPT app, which will allow it not only to better control the shopping experience, but also to access buyer data earlier in the purchase journey, whereas Instant Checkout only gave it visibility at the moment of the final transaction.
Shopify also confirms that the e-commerce experience in ChatGPT will be redesigned. Shopify merchants will remain present in the chatbot, but payment will no longer be native: it will take place on the merchant's online store. Good news for merchants: they will not have to develop a specific app for ChatGPT to benefit from this integration.
The structural tension between AI platforms and retailers
This reversal highlights an underlying tension between two incompatible logics. Platforms like OpenAI and Google want to become the universal interface layer where all commerce happens. Retailers, for their part, have spent decades building their customer relationships and have no intention of becoming mere interchangeable catalogs inside someone else's app.
Amazon perfectly illustrates this defensive stance : the Seattle firm blocked dozens of AI agents, including ChatGPT, from accessing its site, and sued Perplexity for attempting to scrape its pages via its Comet browser. Perplexity responded by accusing Amazon of simply wanting to protect its ad revenues, which AI agents cannot "see". Meanwhile, Amazon continues to invest in its own tools: the shopping chatbot Rufus, the "Buy for Me" agent, and the "Shop Direct" feature allowing browsing of products outside its own catalog.
Google, for its part, is in a more comfortable position. Deeply integrated into shopping via Search and its advertising, hosting third-party shopping agents in Gemini does not threaten its business model. On the contrary, it keeps users within its ecosystem.
A string of announcements that raises questions
The abandonment of Instant Checkout does not occur in a vacuum. As some experts point out, this reversal is part of a series of U-turns by OpenAI : the Stargate project announced at $500 billion with Donald Trump in January 2025, which Bloomberg says is facing significant delays; the partnership with Nvidia announced at $100 billion in September 2025 and quietly reduced to $30 billion in early March 2026; or the deal with the Pentagon urgently announced by Sam Altman, which had to be partially walked back a few days later after widespread criticism.
These repeated episodes are fueling investor worries, especially as OpenAI is considering a stock market listing amid a possible AI bubble. Several large tech companies that went public in 2025 are now trading roughly 60% below their IPO price. Altman himself acknowledged that these recent episodes had been a "good learning experience," a phrase that rings a bit hollow after more than a decade of AI messaging.
The market reaction to the announcement of the abandonment of Instant Checkout is revealing: shares of several specialized platforms—travel booking, meal delivery—jumped 3 to 13% in the following hours. Investors, who had feared that ChatGPT would eat into these services, regained some reassurance.
An industry under construction, not in ruins
That said, it would be excessive to read the death of agentic commerce into the abandonment of Instant Checkout. Emily Pfeiffer of Forrester is clear on this point: " It is not the death of agentic commerce." Everyone thinks others have found the solution or are further along. The reality is that no one has found the formula yet.
Google has just updated its shopping agent platform with capabilities that ChatGPT has not yet mastered: real-time product data uploads to avoid stockouts and pricing errors, adding multiple items to the cart simultaneously, and connection to loyalty programs.
The model that is emerging, " own the agent, rent the distribution ", could well become the industry standard. Retailers with sufficient resources will invest in their own agents, which they will then deploy on AI platforms as additional distribution channels. For others, intermediate solutions, like the one Shopify offers its merchants, will provide visibility without requiring bespoke development.
The article “OpenAI abandons Instant Checkout: what ChatGPT's payment failure reveals” was published on the site Abundance.
