Buy without leaving ChatGPT: the promise that deflated in six months
OpenAI promised checkout without leaving ChatGPT. Six months later, Forrester counts only about thirty Shopify stores wired in, and Walmart calls it a temporary moment. What really matters: being readable when the chatbot advises, not taking payment inside the chat.
On 29 September 2025, OpenAI announced that people would soon be able to buy a product without ever leaving a ChatGPT conversation. One seller on Etsy, then more than a million Shopify stores "coming soon", an assistant that talks with you and, in the same flow, takes the payment. At the time, OpenAI claimed 700 million weekly active users. The message was clear: Google search, the product page, the cart, the whole classic online shopping path, was about to be overtaken by a conversation.
Six months later, Forrester counted how many Shopify stores actually used that built-in checkout feature. Result in February 2026: about thirty. Not a million. About thirty. And on 4 March, Daniel Danker, a vice president at Walmart, publicly called in-chat payment a "very temporary moment". Walmart had nonetheless put about 200,000 products onto that feature as early as November 2025.
That reversal says something important about how AI chatbots are really changing commerce. Not the spectacular version people imagine when they read the announcements. A slower, more technical version, and one that plays out somewhere other than the "pay" button.
Why in-chat payment stalled
Three reasons explain the slowdown, and none of them is anecdotal.
The first is money. Walmart compared two journeys: a customer who buys directly inside the chatbot conversation, and a customer who clicks from the chatbot to the Walmart site to finish the purchase normally. The second journey converts about three times better than the first. At the same time, visitors who arrive on Walmart from ChatGPT, without going through an in-chat purchase, become new customers at roughly twice the rate of a classic Google search. The chatbot brings high-quality traffic. It loses people when it tries to sell to them directly without sending them out of the conversation.
The second reason is data. Forrester observed that the information used by these in-chat purchases (available stock, delivery times, shipping fees) often came from an automatic scrape of the merchant site, a scrape that was not always up to date. A customer is offered a price or a lead time that no longer matches reality. Nothing is worse for trust in a purchase journey.
The third is a question of power. Large online retail platforms are not necessarily keen for an external chatbot to sit between them and their customer, take the payment, and reduce them to a warehouse. Walmart has since added rules that stop some AI agents from taking a user straight to the checkout page. Amazon, for its part, keeps control: its own automated shopping feature, renamed "Alexa for Shopping" in May 2026, is still designed to buy on Amazon, not elsewhere.
OpenAI has not removed its in-chat shopping feature. A spokesperson told the US press that it "is moving into Apps, where purchasing can happen more seamlessly". In plain terms: the project is being parked in a less visible corner, without being announced as a failure, while the approach is reworked. Moved, not buried. But the signal to the market is unambiguous: selling entirely inside the conversation, right away, at scale, was not ready.
What is really happening: search with AI, buy on the site
While one-click payment inside the chat was hitting these obstacles, another, quieter movement kept growing. The numbers show it clearly.
| Metric | Figure | Source |
|---|---|---|
| US visitors who have already bought directly inside an AI tool | 22% | Semrush, December 2025 |
| US visitors who bought after a search done via AI | 50% | Semrush, December 2025 |
| Online transactions involving an AI agent | about 3% | Checkout.com, June 2026 |
| Merchants preparing for this shift | 89% | Checkout.com, June 2026 |
| Rise in AI-referred traffic to US merchant sites (Nov–Dec 2025) | +693% | Adobe |
| Share of US online sales attributed to AI platforms in 2026 | about 1.5% ($20.9 billion) | eMarketer |
These figures tell a coherent story: half of shoppers already use an AI chatbot to research, compare and get advice before buying. But the transaction itself, the one that involves a payment card, still happens mostly on the merchant's site. Analysts have named this pattern "discover in AI, buy on the site". The chatbot becomes a shop-floor assistant that walks you to the till, not the till itself.
That is exactly the dynamic Shopify chose to embrace rather than fight. On 24 March 2026, the platform turned on by default, across 5.6 million stores, a feature that makes its catalogues readable by AI chatbots without necessarily handing them control of payment. Shopify claims, on its own unaudited figures, AI-driven traffic multiplied by seven and AI-attributed orders multiplied by eleven since January 2025. The good numbers are not in sales closed inside a chatbot. They are in the number of customers who land on the site after being convinced by a chatbot.
What this changes for an online store
This course correction does not mean anyone can stop preparing. It simply changes where to put the effort.
There is no need to rush into a payment system built specifically for one chatbot or another: most stores do not need one today, and the format itself is still unstable. OpenAI is reworking its approach, Google is advancing on its side with its own agent payment system, still young and little deployed outside the United States, and Perplexity is testing a buy button with PayPal reserved for its paid subscribers. None of this is stable or widespread in France at the time this article is written.
What does deserve immediate attention is an AI chatbot's ability to understand your catalogue correctly when a customer asks it for advice. That means product pages that give information in clear sentences, not only in a nested technical-spec table that most crawlers ignore. It means stock, price and delivery time that match reality at the moment the chatbot looks them up, because that is precisely the flaw Forrester identified in the first in-chat payment trials. It also means customer reviews and a returns policy that are easy to find, because a chatbot that recommends a product also wants to know whether someone regretted buying it.
Classic Google search has not left the equation. The four studies cited above converge on one point: the signals that made a good merchant site yesterday (load speed, clear structure, complete content) remain the base on which an AI chatbot builds its understanding of a product. The real change is that this content now has to be written and structured for two readers at once: the human who scrolls, and the program that reads the page in a few seconds before deciding whether your product deserves to be recommended.
The takeaway
The story of in-chat payment, launched with fanfare then quietly parked six months later, should make anyone cautious of promises that announce a total, immediate upheaval of online commerce. Reality moves in smaller, measurable steps, often less spectacular than the opening announcement. A chatbot that sends a qualified visitor to your site is worth far more, today, than a chatbot that promises to take payment on your behalf tomorrow.
So the right question is not "should I plug my site into AI payment". It is "if a customer asks ChatGPT, Perplexity or Gemini for advice on a product I sell, does my catalogue give them enough to answer correctly, with the right information, at the right time". That is a readability question before it is a payment-technology question.
Vurto turns an e-commerce catalogue into pages AI chatbots can read and scores each product page on ten clarity criteria, so your catalogue is ready to answer before a customer even asks.