ChatGPT Moves Into Commerce: AI Agents Start Paying, and E-Commerce's Customer Is Becoming a Machine
OpenAI's new ChatGPT Wallet lets AI agents pay at real checkouts, and discovery is moving into chat answers. For traditional e-commerce the contest shifts from capturing human attention to being readable by machines — while bot walls, failed agent purchases and uninterested shoppers show how early it still is.
On November 13, OpenAI gave ChatGPT a wallet. Built with American Express, Mastercard and Visa, ChatGPT Wallet lets the assistant's agents — OpenAI calls them dots — shop on a merchant's own website and pay at its normal checkout with a tokenized card, after the user approves the purchase. The first rollout is narrow: ChatGPT Pro subscribers in the United States, aged 18 and over, with US-issued cards, and no fees announced. It is the moment the world's most popular chatbot stopped only recommending products and started paying for them.
This is OpenAI's second attempt. In September 2025 it launched Instant Checkout, which kept the whole purchase inside the chat window; Walmart then found that in-chat checkout converted about three times lower than sending shoppers to its own site, and OpenAI shut the feature down in March 2026. The new design inverts the old one: discovery happens in the chat, the sale happens on the merchant's site, and the merchant keeps the customer. Days before the wallet, OpenAI also opened Businesses in ChatGPT in beta, letting companies claim their profiles, supply structured product feeds and field branded agents inside the chat — with a Commerce Partner Playbook telling sellers how to prepare.
For traditional e-commerce, the first thing that moves is the entrance. Twenty-five years of online retail have been a contest for human attention: ads, search rankings, storefront design, livestreams. An agent has no attention to capture. It reads product data, prices, stock levels, delivery times and return terms, compares them across sellers, and buys from whichever shop it could actually read. Shopify says orders arriving from AI search in the first quarter of 2026 were already about 13 times the level of a year earlier. The industry has started calling the new pattern B2A — business to agent — and the skill that matters is less persuading a person than being findable, understandable and trustworthy to a machine.
The second change is what wins a sale. An agent is immune to photography, copywriting and brand theatre; it weighs specifications, total price, delivery speed and returns. For routine purchases — detergent, batteries, phone chargers — brand recognition counts for less and logistics counts for more. The third change is friction, and it is still severe: a November test sent an AI shopping agent to 200 retailers in the US and UK and bot protection blocked it at 14.5% of them; in a Visa test the same month, 22 models playing shoppers failed 42.8% of the time. Around the fight, protocols are multiplying — OpenAI and Stripe's Agentic Commerce Protocol against the Universal Commerce Protocol Google and Shopify launched in January — while Amazon, the most resistant of the big retailers, sued Perplexity over its Comet shopping agent before an appeals court overturned the block in August.
The demand side is not there yet. A McKinsey survey of about 21,000 consumers published in October found 53% had used AI for shopping in the previous three months, but only 7% to 18% of American AI users had actually completed a purchase through it, depending on the category. A Salesforce survey published on November 12 found three in four holiday shoppers were not interested in a personal agent buying for them. So the honest picture is a small, clumsy beginning with very large stakes: McKinsey has estimated agents could steer up to a trillion dollars of US retail revenue by 2030. Traditional e-commerce will not vanish, but its shelf is being rearranged for a customer that never sees the shop — and shops that machines cannot read will quietly stop being chosen.
iHolyshit review I have long believed the AI era would produce commerce built for AI, not for people — and this month it stopped being a thought experiment. When the buyer is a machine, storefront decoration, listing copy and livestream patter all lose their power; what decides the sale is whether your data is accurate, your stock is real, your price holds up and your delivery is fast. It feels like e-commerce in 1999: most people think it will come to nothing, while a few are already rearranging their shelves. This time the change is harsher than switching platforms — the customer itself is changing species.
November 18, 2026
Share this article
What do you think of this story?
Anonymous feedback, up to 5 reactions per article.
Reader comments
Comment without signing in, or sign in to display your member name.
Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.
No comments yet. Be the first to share.