Last holiday season, something shifted. Traffic to US retail sites from generative AI tools increased by 693 percent year over year, according to Adobe Analytics. That alone would be a headline. But the more important number is what happened next: those AI-referred shoppers converted 31 percent more than visitors arriving from paid search, email, and every other traditional channel.

AI is no longer just helping people research products. It is becoming where they buy them.

In the space of four months, every major AI platform launched its own checkout experience.

OpenAI moved first. In September 2025, it launched Instant Checkout in ChatGPT, built in partnership with Stripe and powered by the open-source Agentic Commerce Protocol. US users can now buy directly from Etsy sellers and over a million Shopify merchants, including Glossier, SKIMS, and Spanx, without leaving the chat. Walmart, Target, and Salesforce Commerce Cloud merchants followed within weeks.

Google responded at NRF 2026 in January. CEO Sundar Pichai announced that users would soon be able to shop directly within AI Mode and Gemini, paying via Google Pay. Google also unveiled its Universal Commerce Protocol, co-developed with Shopify, Etsy, Walmart, and Target, and endorsed by over 20 companies including American Express, Macy’s, and Visa. Its “Buy for Me” feature goes further, using an AI agent to navigate to a merchant’s website and complete the purchase on your behalf.

Microsoft entered the race at the same conference, launching Copilot Checkout to let users buy within its AI chatbot. Amazon had already introduced “Shop Direct” and its own “Buy for Me” agent, which purchases items from third-party brand sites on a shopper’s behalf.

Four platforms. Four checkout experiences. All launched within the same window.

The scale of this change is worth sitting with. OpenAI says more than 700 million people use ChatGPT each week. Google’s Shopping Graph contains more than 50 billion product listings, with over two billion refreshed every hour. Similarweb data shows ChatGPT holds roughly 68 percent of AI chatbot web traffic, with Google Gemini at 18 percent.

Adobe’s holiday data tells the story of what these audiences do when they arrive at a retail site. AI-referred shoppers were 33 percent less likely to bounce, spent 45 percent more time on site, and viewed 13 percent more pages per visit. AI-driven revenue per visit was up 254 percent during the holiday season. On Thanksgiving, AI conversions were 54 percent higher than non-AI traffic. On Black Friday, 38 percent higher.

These are not research browsers who leave to buy elsewhere. These are high-intent shoppers arriving with a clear sense of what they want because an AI already helped them figure it out.

eMarketer projects 63.3 million US consumers will use AI platforms and assistants to shop this year. By 2030, AI-powered tools and agentic commerce could represent a $3 trillion to $5 trillion opportunity globally.

Underneath the consumer-facing features, a quieter but more consequential battle is playing out: who defines the standard for how AI agents transact?

OpenAI and Stripe built the Agentic Commerce Protocol, or ACP. It is open-source, and OpenAI has published its documentation for any merchant or developer to build against. If a merchant already processes payments with Stripe, they can enable agentic payments in as little as one line of code. If they use another processor, Stripe offers a Shared Payment Token API to participate without switching.

Google countered with its Universal Commerce Protocol, or UCP. Also positioned as an open standard, but co-developed with a different coalition of retailers and payment networks.

This is where it gets interesting for payments professionals. We now have two competing open protocols, each backed by a dominant AI platform and a different set of commerce partners, both claiming to be the standard for how AI agents discover, negotiate, and complete purchases.

The history of payments is littered with protocol wars. EMV took decades to settle. Real-time payment schemes still lack global interoperability. The question is whether AI commerce will repeat that pattern or whether market dynamics will force consolidation faster. With Stripe deeply embedded in ACP and Google Pay anchoring UCP, the answer may depend on which payment rails merchants already use.

For anyone building or operating in payments infrastructure, several implications stand out.

Authentication is being rearchitected. When an AI agent completes a purchase on your behalf, who is authenticating? Google’s “Buy for Me” navigates to a merchant’s site and checks out using Google Pay. OpenAI’s Instant Checkout passes payment tokens between buyer and merchant. Neither of these flows maps cleanly onto existing 3D Secure or Strong Customer Authentication frameworks. The regulatory implications have barely been explored.

The merchant of record question is getting complicated. Every platform is careful to state that the retailer remains merchant of record. But when an AI agent selects the product, negotiates the price through “Direct Offers” ad units, and processes the payment through a platform-controlled protocol, the practical reality of that merchant relationship starts to thin out. Brands are paying attention: how they show up in AI-generated results is becoming as important as their website.

Fraud models need updating. AI-referred traffic behaves differently. It converts higher and bounces less, but the transaction patterns are unfamiliar to fraud engines trained on years of browser-based shopping behaviour. When an AI agent fills in shipping details and selects a payment method on a user’s behalf, the signals that fraud systems rely on, such as typing cadence, mouse movement, and session history, simply do not exist.

Interchange economics could shift. If AI checkout becomes a meaningful share of e-commerce transactions, the platforms hosting those checkouts gain enormous leverage. OpenAI charges merchants a “small fee” on completed purchases. Google is integrating advertising into AI Mode through Direct Offers. Both models create new intermediaries sitting between the card network and the consumer, and history tells us intermediaries with scale tend to negotiate.

The early winners are clear: Shopify and Stripe. Shopify merchants are integrated across ChatGPT, Google, and Microsoft. Stripe powers the payment infrastructure for OpenAI’s checkout and offers the bridge for non-Stripe merchants to participate. Their bet on being the connective tissue for commerce, rather than the destination, is paying off.

The position is more complicated for large retailers. Walmart and Target signed up early to Google’s UCP, and both are available through ChatGPT. But participation means ceding some control over the shopping experience to an AI intermediary. Forrester analyst Sucharita Kodali has questioned whether chat-based commerce solves a real problem for retailers, noting that “e-commerce isn’t a problem that needs to be fixed.”

She may be right in the short term. eMarketer projects AI-driven e-commerce sales will account for just 1.5 percent of overall online shopping in 2026. But the trajectory matters more than the current share. AI traffic to retail sites has grown from negligible to converting better than paid search in roughly 18 months.

The biggest losers may be traditional search advertising. If consumers ask ChatGPT for “the best running shoes under $100” and buy within the chat, Google loses a search query and the ad revenue attached to it. Google clearly sees this threat, which explains why it is moving so aggressively to embed checkout into its own AI experiences.

We are watching something that does not happen often in commerce: a genuine platform shift in how consumers discover and buy products. The last comparable moment was the move to mobile, which took the better part of a decade to fully play out.

This one could move faster. The infrastructure is already being built. The protocols are being written. The payment rails are being connected. And unlike mobile, which required hardware adoption, AI checkout runs on the devices people already have.

The open questions are significant. Will ACP and UCP converge or fragment? How will regulators respond to AI agents making purchasing decisions? What happens to brand equity when an AI decides which product to recommend? And who bears liability when an AI agent buys the wrong thing?

These are not hypothetical problems. They are design decisions being made right now, in real time, by the companies building the next generation of commerce.

Related reading from Major Matters:

Four platforms are racing to own the AI buy button. Which protocol wins, and what does that mean for the merchants and payment companies caught in the middle? We would love to hear your take.