OpenAI just walked back its native checkout feature after roughly a dozen Shopify merchants went live. Google, meanwhile, is doubling down on the same premise. At Google Marketing Live this week, the company announced expanded checkout integration through its Universal Commerce Protocol, letting users complete purchases without leaving AI chat, YouTube ads, or Google Maps. The contrast is almost too convenient: one platform retreats, the other advances.
For B2B marketing executives watching this unfold, the question isn't whether agentic commerce will matter. Morgan Stanley projects $190 billion to $385 billion in U.S. e-commerce spending through agents by 2030. The question is whether the infrastructure exists to capture that value today, and what the OpenAI retreat tells us about the gap between vision and execution.
The Protocol War Nobody Asked For
Google's Universal Commerce Protocol, launched in January 2026, was co-developed with Shopify, Etsy, Wayfair, Target, and Walmart. It establishes a common language for AI agents to interact with merchant systems across discovery, checkout, and post-purchase support. The protocol is designed to work with existing industry standards like Agent2Agent and Model Context Protocol, positioning it as infrastructure rather than a walled garden.
OpenAI's competing Agentic Commerce Protocol, built with Stripe, launched in September 2025 with similar ambitions. PayPal joined in October. The vision was identical: collapse the funnel from discovery to purchase inside a single conversational interface.
Then reality intervened. Forrester reported in March that OpenAI scaled back Instant Checkout after finding that users researched products in ChatGPT but didn't complete purchases there. Modern Retail confirmed that Etsy saw disappointing sales volume from the feature. Shopify's president acknowledged at an investor conference that only about a dozen of its millions of merchants had gone live with any agentic shopping feature.
Google's response at GML 2026 was to accelerate. The company announced a Universal Cart that works across Search, YouTube, Gmail, and the Gemini app. Merchants on UCP can now export loyalty points and exclusive discounts directly into agentic ads. The checkout integration extends to YouTube video ads and Google Maps, meaning someone watching a video or searching for directions can theoretically complete a purchase without leaving the surface.
The Adoption Gap That Matters
The statistics paint a contradictory picture. MetaRouter's analysis shows 39% of consumers already use AI for product discovery, with an 805% year-over-year increase in AI traffic to retail sites on Black Friday 2025. Yet ChatGPT referrals convert 86% worse than affiliate links, and less than 0.2% of e-commerce sessions currently come from ChatGPT.
First Page Sage's May 2026 research found that enterprise agentic AI adoption sits at 25%, with mid-market and SMB companies growing faster due to turnkey solutions. But adoption doesn't equal production deployment. The same research shows that 88% of AI agents fail to reach production, though the survivors deliver an average 171% ROI.
For marketing executives, this creates a familiar forecasting problem. The demand signal is real. The conversion infrastructure isn't. And the two competing protocols mean merchants face a choice about which ecosystem to prioritize, or whether to implement both.
What Google's Bet Reveals About Attribution
The GML 2026 announcements included something more interesting than checkout features: new measurement tools designed for AI-mediated commerce. Google is launching incrementality experiments across campaign types with lower budget requirements, plus a reimagined Google Analytics 360 as a "command center for modern measurement."
This matters because the core attribution problem in agentic commerce is visibility. In traditional e-commerce, you see impressions, clicks, dwell time, add-to-cart events, and funnel drop-offs. In agent-mediated commerce, the behavioral data stream starts at the add-to-cart moment. Discovery, browsing, consideration, and preference refinement all happen inside the AI interface. Your attribution model goes dark precisely where the decision is made.
Google's approach is to own both the agent surface and the measurement layer. If checkout happens inside AI Mode or YouTube, Google captures the full journey. The Universal Cart becomes a data asset as much as a conversion tool.
For B2B marketers accustomed to multi-touch attribution debates, this is a familiar dynamic with higher stakes. The platform that controls the agent controls the signal. And the signal determines budget allocation.
The CFO Question: Where's the Payback?
Let me model this plainly. Mordor Intelligence estimates the agentic AI retail market at $60.43 billion in 2026, growing to $218.37 billion by 2031 at a 29.29% CAGR. That's the addressable opportunity.
Against that, consider the implementation cost. UCP integration requires structured product data, real-time inventory feeds, and checkout API connections. Merchants who want to participate in both Google's UCP and OpenAI's ACP need dual implementations. One analysis suggests dual implementation captures 40% more agentic traffic, but the combined take rate for ChatGPT plus Shopify runs around 9.2%, with OpenAI charging a 4% transaction fee.
The payback calculation depends on whether agentic commerce cannibalizes existing channels or creates incremental demand. If a customer who would have searched on Google and clicked through to your site instead completes the purchase inside AI Mode, you've traded a lower-cost conversion for a higher-cost one with less data visibility. If the agent surfaces your product to someone who wouldn't have found you otherwise, that's genuine lift.
OpenAI's retreat suggests the cannibalization scenario dominated their early results. Users researched in ChatGPT, then bought elsewhere. Google's bet is that owning more surfaces, from Search to YouTube to Maps, creates enough touchpoints to capture the conversion where it naturally occurs.
A Two-Week Pilot Design
If you're evaluating agentic commerce readiness, here's a structured approach:
Week one: Audit your product data infrastructure. UCP requires structured feeds with real-time inventory, pricing, and fulfillment data. Map your current Merchant Center setup against UCP's catalog, cart, and checkout capabilities. Identify gaps in variant resolution, dynamic pricing, and loyalty program integration.
Week two: Run a controlled test on a single product category. Enable UCP checkout for a subset of SKUs with clean data. Measure conversion rate against your baseline Google Shopping performance. Track the attribution gap: what percentage of UCP-initiated sessions can you connect to downstream behavior?
The risks are straightforward. Protocol fragmentation means your investment in one ecosystem may not transfer to another. The measurement gap means you're flying partially blind on incrementality. And the take rates mean your margin structure needs to accommodate platform fees that didn't exist in direct channels.
What This Means for Pipeline
Google's GML 2026 message was explicit: "The only way to win in the age of AI is with AI." That's marketing language, but the underlying bet is structural. Google is positioning itself as the infrastructure layer for agentic commerce, not just an advertising platform.
For B2B marketing executives, the implication is that your pipeline models need a new input. Agentic commerce isn't a channel in the traditional sense. It's a layer that sits between intent and conversion, potentially across every surface where your customers interact with AI. The companies that build the data infrastructure to participate will have optionality. The companies that wait for the protocols to consolidate may find the standards set without them.
OpenAI's retreat doesn't mean agentic commerce is a mirage. It means the execution is harder than the vision. Google's acceleration doesn't mean they've solved it. It means they're willing to spend the next several years finding out.
