OpenAI Pulls Back on ChatGPT Instant Checkout
After six months of testing, OpenAI has decided to abandon its Instant Checkout feature inside ChatGPT, ending the ambition to turn the AI chatbot into a direct shopping platform. The company confirmed the shift in a public statement, saying purchases will now route through merchant apps like Instacart, Target, Expedia, and Booking.com.
Out of Shopify's millions of merchants, only about 12 went live with ChatGPT checkout during the entire testing period. OpenAI never built infrastructure to collect and remit state sales taxes, signaling transaction volumes never reached a meaningful scale.
What Instant Checkout Was Supposed to Do
When OpenAI launched Instant Checkout in September 2025, the goal was straightforward. A user searching for a product inside ChatGPT would see options to purchase directly within the chat, without visiting any external website. The company described ChatGPT as acting like a “digital personal shopper,” securely passing payment and shipping details between the user and the merchant.
Partners like Shopify and Etsy were brought in as aggregators to speed up merchant adoption. Users would pay with a card already saved in ChatGPT or add a new one. OpenAI collected a commission on completed purchases, while merchants handled fulfillment on their end. The feature launched exclusively in the United States, with plans to expand to other markets later.
Three Reasons the Feature Failed to Scale
Despite a strong premise, the project ran into structural problems from the start. Shopify president Harley Finkelstein publicly stated at an investor conference that the bottleneck was on OpenAI's side, not the merchants'. Merchants were willing to participate, but OpenAI had not built the systems needed to bring them on board at scale.
Merchant adoption stalled
Getting merchants onto the Agentic Commerce Protocol (ACP) required them to build and maintain structured product feeds, integrate with new catalog and payment APIs, and keep pricing, inventory, and shipping data synchronized in real time. Large retailers with dedicated engineering teams took months to complete this work. Smaller merchants, who make up the majority of online stores, lacked the resources to justify the effort when agent-driven transaction volumes were still negligible.
Product data was hard to keep accurate
Merchant product listings, including prices, stock levels, and shipping options, need constant updates to stay accurate. Any system relying on pre-ingested catalog data risks showing outdated information. Prices change. Items go out of stock. Shipping costs vary by location. OpenAI employees found keeping thousands of product listings from potentially thousands of merchants synchronized inside ChatGPT was far more difficult than anticipated.
Fraud safeguards were not in place
Payment companies raised concerns about AI initiating fraudulent or erroneous transactions. Protocol-based systems address this by requiring explicit merchant consent, but this approach leaves the vast majority of merchants, those who never opted in, completely inaccessible to AI agents. OpenAI had not resolved this problem during the testing period.
How Users Actually Behaved
Internal findings from the testing period revealed a clear pattern. Users actively researched and compared products inside ChatGPT, but they did not complete purchases there. When the time came to buy, they returned to familiar ecommerce platforms. The chat interface served as a discovery tool, not a checkout destination.
Forrester research confirms this pattern. Among US, UK, and Canadian adults who regularly use AI answer engines, completing a purchase inside the platform ranks as their least-adopted use case. Asking general questions and researching products rank first and second.
This behavior points to a trust gap. Consumers are comfortable using AI to research options, but they want to finalize payments in environments they already know and trust. Closing this gap takes time, and OpenAI concluded that the current setup would not get there fast enough.
What OpenAI Is Doing Instead
OpenAI is not walking away from commerce entirely. The company will continue developing ACP alongside Stripe and PayPal, but the scope is now narrower. The focus shifts to large retailers building dedicated apps inside ChatGPT, including Instacart, Target, Expedia, and Booking.com. These partnerships cover a small number of high-volume merchants rather than the broad base of online stores OpenAI originally targeted.
For ChatGPT, the practical role becomes a referral engine. The platform guides users through product discovery and comparison, then sends them to merchant apps or websites to complete the purchase. OpenAI earns commissions on traffic generated, not on transactions processed.
What This Means for AI Commerce
The retreat from Instant Checkout does not mean AI-assisted shopping is finished. One specific approach did not work at the scale OpenAI needed. The protocol-dependent model, which requires merchants to opt in, maintain data feeds, and build integrations, works for a small number of large retailers. The model does not extend to the millions of smaller merchants making up the long tail of ecommerce.
The travel sector responded to the news with notable optimism. Expedia shares jumped over 12%, Booking Holdings gained around 8%, and Tripadvisor rose approximately 5% following the announcement. Investors read the move as confirmation that these platforms retain control over their booking flows, with AI serving as a source of customer referrals rather than a competitor processing transactions directly.
For the broader ecommerce market, the question now is what infrastructure fills the gap. Universal checkout tools, which work across any merchant site without requiring protocol adoption, address the exact problems OpenAI encountered. Whether those tools gain traction will shape how AI agents handle purchases over the next few years.
