OpenAI has pulled back from letting shoppers pay in ChatGPT. Instead of completing a purchase in the chat window, users will now finish orders in retailer apps connected to ChatGPT, as reported by The Information and covered by Search Engine Land.
Instant Checkout, once pitched as a big commerce play, is moving to Apps. An OpenAI spokesperson confirmed that purchases will happen in connected services and not natively in ChatGPT. The company will prioritise product search and discovery in ChatGPT and continue working with Stripe on the Agentic Commerce Protocol to support app based transactions.
The rethink comes after OpenAI found that people research products in ChatGPT but do not complete purchases there. Only a small number of merchants were actively using native ChatGPT checkout. Shopify president Harley Finkelstein said that only about a dozen Shopify merchants were using AI tools, even though Shopify supports integrations with ChatGPT, Gemini and Copilot. Against Shopify’s overall merchant base, that is tiny.
For a product once described by Bernstein analysts as the start of “Agentic One”, the change is telling.
Is This A Setback For Agentic Commerce?
Chris Jones, Managing Director at PSE Consulting, does not think so.
“The news that OpenAI has scaled back Instant Checkout should not be seen as a failure of agentic commerce. Instead, it reflects the ambition of integrating autonomous AI agents into real world shopping at scale. These systems must navigate a vast and complex landscape including product data, identity, fraud, tax, liability, loyalty and payments across thousands of merchants.”
That list is long for a reason. Shopping looks easy at first glance. Underneath, it runs on messy systems.
Jones spoke on structural friction identified in the recent Adyen report on agentic commerce. “The recent Adyen report on agentic commerce highlights the same structural challenges. Based on insights from more than 200 enterprise merchants and direct work with AI commerce platforms, it identifies five constraints that go beyond any one company’s control: protocol fragmentation, product data not designed for machine consumption, legacy enterprise stacks, trust and liability frameworks, and onboarding at scale. Together these factors mean that what works in demos can break quickly in production environments.”
A smooth demo is not the same as a checkout that works across thousands of retailers, currencies and tax regimes.
Leigh McKenzie, director of online visibility at Semrush, told Search Engine Land that two forces are slowing agentic commerce: infrastructure and trust. Real time catalogue normalisation across tens of millions of SKUs is a decade scale problem that Google solved with Merchant Center. Consumers also default to checkout flows they trust such as Apple Pay, Google Wallet and Amazon one click.
More from News
- How Much Internet Freedom Do Different Countries Actually Have In 2026?
- Ofcom’s Wholesale Pricing Rules Risk Creating A UK Broadband Duopoly
- Could China Really Switch Off Nigeria’s Satellites? The $11.44M Dispute Explained
- Your Food Delivery Services May Be Impacted By The Global Oil Crisis, Here’s How
- From 19 March, Debit And Credit Card Users Can Control Their Contactless Limits In The UK
- Why Are Anti Drone Patents Up 27% In A Year?
- Boost Named monday.com’s Best Professional Services Partner In EMEA For The Second Consecutive Year
- Tech Experts Discuss The Future Of Agentic AI After Agent Mines Crypto Without Receiving Instructions
Are Payments The Real Problem?
Payments are only a portion of the issue here.
“From a payments perspective, some challenges remain addressable. Fraud, tax and payment flows can be improved through collaboration across payment networks, infrastructure providers and AI platforms. These are areas where scale, standards and partnerships can drive progress over time,” Jones said.
He is direct about the limits of a narrow view.
“For payments specialists, the takeaway is clear. Agentic commerce success requires a holistic approach. Solving payments in isolation is not enough. Systems must handle multiple protocols, maintain accurate product data, support persistent identity, manage liability and scale onboarding. The recent developments from ChatGPT and the insights from Adyen should not be seen as a retreat, but as a reminder that the industry is still building the infrastructure needed to make autonomous shopping safe, reliable and seamless.”
Checkout in a chatbot made for a catchy headline. The hard work sits deeper in the stack.
What Does This Mean For Retailers, Especially Fashion?
Style Arcade reports that during Black Friday, $14.2 billion in global online sales and $3 billion in US online sales were driven by AI agents. Discovery works. Conversion is harder.
ChatGPT can recommend a dress. It can compare colours and prices. The trouble starts when a customer asks about size availability or return rules. Style Arcade’s data shows the average fashion retailer holds 18% excess inventory in slower moving sizes while missing 22% of demand in faster sellers. That is not a chatbot issue. That is a buying and stock allocation issue.
Rent the Runway reports the fastest shift to smaller sizes in 15 years. Lululemon flagged missed opportunities in smaller sizes in its 2024 earnings. These are operational gaps. No embedded checkout button fixes a broken size curve.
OpenAI cited operational complexity, fraud prevention and customer disputes in its decision. That lines up neatly with what consultants and retail data have been saying. Agentic commerce is not dead. Perhaps it is just now running into the real mechanics of retail.
ChatGPT will keep guiding shoppers through discovery. The final click, at least for now, belongs to retailer apps.
