For the past year, the conversation around AI and ecommerce has centred on discovery: whether ChatGPT recommends your product when a customer asks for one. That conversation is already becoming outdated. The infrastructure for AI agents to complete the purchase themselves, without the customer ever landing on a product page, is now live and being adopted by some of the largest retailers in the world.
Google has launched agentic checkout across Search’s AI Mode and Gemini, with a “Buy for me” feature that lets an AI agent execute a purchase directly on a merchant’s site rather than simply pointing a customer toward it. Wayfair, Chewy and Etsy are among the early retail partners already live with the feature.
At NRF 2026, Google went further, unveiling the Universal Commerce Protocol alongside Shopify, Etsy, Wayfair and Target, a standard designed to cover the entire commerce journey from discovery through to post-purchase support, backed by more than twenty payment and retail partners.
Meanwhile, ChatGPT Shopping is now live for all United States users, connected to Etsy and more than a million Shopify merchants. These are not pilot programmes tucked away in a lab. They are default features being rolled out to hundreds of millions of users.
The card networks have moved just as quickly. Mastercard has completed its first live agentic transactions in markets including Singapore and South Korea, Visa has commercially launched its Trusted Agent Protocol after piloting with more than a hundred partners, and American Express has released purchase protection specifically for AI agent purchases. When payment infrastructure this large starts building for agent-initiated transactions, it signals that the shift is being treated as permanent rather than experimental.
Where the model is still finding its feet
The most useful data point so far isn’t a success story. It’s a stumble. OpenAI initially built Instant Checkout directly into ChatGPT, allowing a purchase to be completed inside the conversation itself. Conversion rates for purchases completed inside ChatGPT ran three times lower than those that redirected to the merchant’s own website, and by March 2026, OpenAI had deprecated the feature in favour of a model where the agent handles discovery and comparison, then hands the shopper off to the retailer’s own site to complete the transaction.
This all complicates the more dramatic version of the “zero-click commerce” narrative. Shoppers appear willing to let an AI agent do the searching, comparing and shortlisting on their behalf. They are, for now, less willing to hand over the final act of paying without seeing where their money is going.
Industry data reflects this pattern: AI adoption in commerce remains concentrated in the early stages of the customer journey, with usage reaching approximately 62% for product comparison, compared with about 23% at checkout and 19% during post-purchase activities.
Agents are becoming the layer that decides which four or five products make the shortlist. Humans are, at least for now, still closing the loop.
Where should an ecommerce business focus first? The immediate opportunity is not rebuilding checkout for autonomous agents. It’s making sure the product is in the shortlist an agent produces when it does the comparing, since that’s the stage where the decision about what a customer even sees is increasingly being made without them.
The numbers retailers are already seeing
Where agent-driven shopping has taken hold, the results are difficult to ignore. Adobe Analytics found that AI-referred visitors had a 38% higher purchase completion rate compared with traditional search visitors during Black Friday 2025, and Salesforce reported that retailers with AI agent integration saw roughly seven times better sales growth during Cyber Week 2025 than those without it. A shopper who arrives via an AI agent has typically already been through a comparison process before they land on the site, which likely explains why they convert at a higher rate than someone still in the early stages of browsing.
The scale of what’s being built toward is significant too. McKinsey estimates that agentic AI will influence three to five trillion dollars in global retail commerce by 2030, and Morgan Stanley predicts that nearly half of online shoppers will use AI shopping agents by then, accounting for roughly a quarter of their spending. Whatever the exact trajectory turns out to be, the direction is consistent across every major forecast: a meaningful share of future revenue will pass through an agent before it reaches a retailer’s own storefront.
What being “agent-ready” actually requires
The practical challenge is that most ecommerce infrastructure was never built for this kind of traffic. Most enterprise commerce stacks were designed around session-based human interaction, where intent enters through search, results are ranked, and users evaluate product pages one click at a time.
An AI agent doesn’t browse that way. It parses a user’s request into structured intent and queries live systems directly, checking inventory for a specific size, pricing endpoints for products under a set budget, and shipping estimators for a delivery deadline, often through protocols like MCP that allow it to query a retailer’s actual data rather than scraping a rendered page. A product catalogue full of inconsistent sizing fields, outdated stock counts, or descriptions written purely for human persuasion gives an agent very little to work with, and an agent that can’t get a clear answer tends to move on to a competitor that can.
A useful starting point for any ecommerce business is a simple audit. Check whether product data is complete and accurate, including structured attributes, honest descriptions, current pricing and real-time inventory. Then test visibility directly by asking ChatGPT or Google’s AI Mode to find products in your category, and see whether your store appears in the results.
Finally, confirm whether your ecommerce platform supports the emerging protocol standards, or at minimum has API coverage for carts, checkout and orders. None of this requires a full platform rebuild. It does require treating structured, accurate product data as core infrastructure rather than a secondary concern handled once a listing goes live.
The shift in what “winning” a sale means
Sabu Thomas, writing on the infrastructure shift underway, frames it plainly: the future of commerce is not just mobile or conversational, it is agentic. The practical implication is that a retailer competing for a sale is no longer only competing for a human’s attention through design, copy and imagery. It’s also competing to be legible and selectable to a system that is evaluating options based on clarity, accuracy and ease of execution rather than persuasion.
A beautifully designed product page still matters for the customer who arrives directly. It does very little for the agent deciding, on a shopper’s behalf, which three retailers even make it into the comparison.
This doesn’t mean every ecommerce business needs to have agentic checkout live tomorrow. Full autonomous purchasing is still finding its footing, as OpenAI’s own retreat from in-chat checkout shows.
What it does mean is that the businesses treating their product data, inventory feeds and structured attributes as a genuine priority now are the ones positioned to be visible when agent-driven shopping moves from roughly a quarter of checkout activity to something considerably larger, which every major forecast suggests is a matter of when rather than if.