AI Can Now Complete the Purchase. Is Your Checkout Ready?

AI Can Now Complete the Purchase. Is Your Checkout Ready?

AI has already changed how people research products. The next change is more significant: AI can now complete the purchase.

On 13 November 2026, OpenAI introduced ChatGPT Wallet, initially for eligible ChatGPT Pro users in the United States. Customers can add a card, ask ChatGPT to buy something and approve the transaction. The card details are tokenised, while the business remains responsible for pricing, fulfilment, returns and customer service.

This is not an isolated experiment. Mastercard has introduced services designed to help merchants and banks recognise legitimate AI-initiated transactions. Visa is already running a live pilot in which authorised AI agents complete approved invoice payments through supplier portals.

The buying journey is moving beyond AI-assisted discovery. Businesses now need to prepare for AI-assisted checkout.

AI commerce creates a different kind of customer journey

A traditional online purchase involves a person visiting a website, browsing products, adding an item to a cart and entering their payment information.

An AI-led purchase can work differently. The customer tells an assistant what they need, sets requirements and approves the final choice. The assistant may then navigate the website, select the right option and complete the transaction.

The customer is still making the decision. The AI is carrying out more of the process.

That changes where friction appears. A confusing navigation menu may matter less if the agent can access the product directly. Inaccurate stock information, unclear delivery terms or a bot protection system that blocks the agent may matter much more.

Businesses need to consider whether their buying journey works for software acting with a customer’s permission, as well as for the customer themselves.

Your product data is becoming part of the checkout

Product information has always influenced sales. In AI commerce, that information becomes operational infrastructure.

An AI assistant needs reliable data before it can make a suitable choice. That includes the current price, available variants, stock status, delivery options, geographic restrictions and returns policy.

If those details conflict across your website, product feed and checkout, the agent may select the wrong item or fail to finish the purchase. A promotional price that disappears in the cart could cause the transaction to stop. A product marked as available when it cannot ship to the customer’s location could create a service problem after payment.

OpenAI’s commerce documentation advises businesses to maintain accurate product feeds and ensure that prices and availability are current. Mastercard similarly describes a journey in which agents discover merchant-provided information, confirm taxes and shipping, and then complete the transaction.

Treat product data as part of the customer experience. Give someone clear ownership of it. Check how quickly changes reach every channel, and test whether the information shown before purchase matches what appears at checkout.

Your website may be blocking legitimate buyers

Most ecommerce websites use bot detection, CAPTCHAs, firewalls and other controls to stop fraud or scraping. Those protections remain necessary, but they can also block authorised AI agents.

This creates a difficult balance. Opening the door to every automated visitor is dangerous. Blocking every automated visitor could reject legitimate customers.

The industry is starting to develop ways to distinguish trusted agents from malicious bots. Visa’s recent B2B payment pilot uses an agent protocol and directory to authenticate an authorised agent before allowing it to enter a supplier portal. OpenAI also advises businesses to review their content delivery network and firewall settings so supported agents can reach the purchase path. 

Ask your ecommerce, security or hosting team what currently happens when a recognised AI agent accesses the site. Review bot rules, blocked sessions and abandoned transactions. You may find that a security control is quietly turning away a valid purchase.

Payment systems will need more context

An AI-led transaction can look unusual to a fraud system.

The buyer may not have followed a normal browsing pattern. The payment could arrive through a token rather than a familiar checkout session. An agent booking a trip might make several purchases across different merchants within minutes.

Those actions could trigger rules designed for human behaviour.

Mastercard announced in September 2026 that it was testing a probability score to help identify AI-initiated transactions. Its aim is to give banks and merchants more context when deciding whether to approve a purchase or request additional checks.

Businesses should speak with their payment provider about how agent transactions are identified and assessed. Monitor decline reasons rather than looking only at the overall approval rate. A rise in rejected agent payments could appear to be weak demand when the real issue sits inside the risk system.

Marketing attribution may become harder

Agent-led commerce also creates a measurement challenge.

A customer might discover a product through an advert, research it through an AI assistant and ask that assistant to buy it days later. The final website session may contain little of the browsing behaviour that marketing teams normally use for attribution.

If your reporting credits only the last visible interaction, the channels that created demand could disappear from the story. At the same time, treating the AI transaction as an entirely new acquisition source could lead to double counting.

Marketing leaders should map how campaign identifiers, referral information and customer consent move through an agent-assisted purchase. Where possible, separate product discovery, recommendation and transaction completion in your reporting.

The goal is not to force every journey into an old attribution model. It is to understand which marketing activities created the customer’s intent and which system completed the order.

Returns and customer service still belong to your business

AI may place the order, but it does not remove the merchant’s responsibilities.

OpenAI states that businesses using ChatGPT Wallet retain the customer relationship and continue to control fulfillment, returns and service. That makes the handover after checkout critical.

Customers need a clear order confirmation, a way to identify the purchase and an obvious route for help. Your service team needs to know whether the order was agent-assisted and what the customer approved.

Review the full post-purchase journey. Check whether confirmation emails contain everything the customer needs, whether support teams can find the order and whether cancellation and refund rules work when an agent initiates the transaction.

Test the complete journey before demand arrives

Do not start with a major AI commerce project. Start with a practical readiness test.

Choose a few popular or complex products and follow the entire journey: finding the correct item, choosing a variant, confirming stock, calculating shipping, applying a promotion, paying, receiving confirmation and requesting a return.

Look for unclear information, inconsistent totals, blocked access and missing order details. Then repeat the test when prices, availability or delivery conditions change.

AI commerce will not remove the need for good marketing or a strong customer experience. It will expose gaps between what your marketing promises and what your systems can deliver.

The businesses that prepare now will make it easier for both people and their chosen assistants to buy with confidence.

If AI is changing how customers discover and buy from your business, talk to The Go! Agency about building a clearer marketing strategy around the complete customer journey.

 

×
✕