Visa plugs its payment network into ChatGPT so AI agents can shop on their own

Visa has integrated its payment infrastructure with OpenAI's ChatGPT, allowing AI agents to recommend products and complete financial transactions on a user's behalf. The final stages of the retail sales funnel, from choosing a product to paying for it, move entirely to autonomous agents, with no human involved at the moment of purchase.

How buying through an AI agent will work

As described, autonomous agents will process user requests, analyze product catalogs from partner merchants and complete the order using Visa's payment rails at any merchant that supports the feature. Payment rails are the technical infrastructure that carries a transaction from the buyer's card to the merchant's account. Integration with ChatGPT means the agent can initiate and complete such a transaction directly in the conversation with the user, without redirecting to the store's website or filling in a payment form by hand.

Why payment networks want agentic commerce

This is part of a broader push by the payments industry and AI developers toward so-called agentic commerce, where an AI agent sits between what the user wants ("find me...", "buy...") and the final purchase. For payment networks like Visa, the stakes run in both directions. On one hand, agentic commerce opens a new transaction channel and keeps Visa as a mandatory link in the payment chain even when the user is talking to a chatbot rather than a store's website. On the other, handing the final purchase decision to an autonomous agent requires new safeguards against errors and fraud, since a human no longer confirms each step manually.

Key elements of the integration:

• Partners: Visa (payment network) and OpenAI (developer of ChatGPT).
• Functionality: product recommendations and purchase completion by an AI agent.
• Human involvement is removed from the final stages of the retail funnel.
• Transactions run over Visa's payment rails at supporting merchants.
• The agent handles the request, evaluates the catalog and completes checkout on its own.

What this changes for retail

For merchants and brands, the integration means adapting their catalogs and checkout systems to deal with a software agent rather than a person. That means providing machine-readable descriptions of products, prices and delivery terms that an AI system can understand. It accelerates a trend of recent years, in which major technology and financial companies compete to become the infrastructure layer of the future "agentic economy": whoever first builds a convenient, reliable bridge between conversational AI and real money will control a huge share of future online purchases.

Regulatory questions are growing at the same time: who is liable if the agent picks the wrong product or makes an unwanted purchase, how the user's consent to a specific transaction is verified, and how card data is protected when passed through a conversational interface. The industry is only now developing standards for agentic commerce, and the Visa-ChatGPT integration is one of the first major precedents through which these rules will likely take shape in practice.

Integrations like this between payment networks and conversational AI assistants fit into a broader shift in e-commerce. For years, retailers invested in making checkout easier for humans: fewer clicks, card autofill, one-click checkout. Now that logic extends to a non-human buyer.

An agent acting on a user's instructions needs a path to completing the deal that is as fast and reliable as the one for a person clicking "buy" on a website. For Visa, as the operator of the payment infrastructure, this means building new protocols to verify that a trusted agent, not a malicious script, is initiating a transaction on behalf of the cardholder. How quickly this model of shopping catches on among ordinary users will depend largely on how transparent and predictable the process feels to the buyer.

Willingness to delegate the final decision to spend money to a software agent requires trust, and the industry still has to earn it.