The Checkout Wars
Who controls the buy button when AI goes shopping?
Three weeks ago, at the National Retail Federation conference in New York, the contours of a coming war became visible. Google unveiled its Universal Commerce Protocol, an open standard for AI agents to handle shopping from discovery through checkout. Microsoft announced Copilot Checkout, letting users complete purchases without leaving its chatbot. Both companies positioned themselves as neutral infrastructure, promising retailers they would remain the merchant of record.
The timing was deliberate. Adobe had just reported that traffic to retail sites from generative AI surged 693% during the 2025 holiday season. The question is no longer whether AI will reshape commerce, but who will control the interface where purchasing decisions actually happen.
The answer depends on which model prevails: consent-based partnerships or aggressive extraction.
The Protocol Play
Google’s UCP represents the most ambitious attempt to standardize agentic commerce. Co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by more than 20 additional partners including Visa, Mastercard, and Stripe, the protocol establishes what Google calls a common language for AI agents and commerce systems to operate together.
The pitch is interoperability. Instead of requiring unique connections for every individual AI agent, UCP allows any compliant agent to interact with any compliant merchant. It supports existing industry protocols including Anthropic’s Model Context Protocol, which has emerged as a de facto standard after being donated to the Linux Foundation’s Agentic AI Foundation.
Microsoft’s approach is narrower but more immediately deployable. Copilot Checkout integrates with PayPal, Shopify, Stripe, and Etsy, letting users purchase from brands including Urban Outfitters, Anthropologie, and Ashley Furniture directly within the Copilot interface. The company emphasizes that retailers remain the merchant of record, handling fulfillment and customer service while Microsoft controls the interface.
OpenAI, meanwhile, has struggled to scale its Instant Checkout feature. Launched in September with Etsy sellers and select Shopify merchants, the service has faced challenges wrangling product data. According to The Information, the in-app checkout feature that OpenAI promised would connect millions of merchants remains largely unavailable. Converting ChatGPT’s 50 million daily shopping queries into actual transactions has proven more difficult than anticipated; real-time synchronization of product information across inconsistent merchant systems is, it turns out, genuinely hard.
The Extraction Model
And then there is Amazon.
In February 2025, the company launched “Buy for Me,” an agentic tool that lets consumers purchase products from third-party websites without leaving Amazon’s app. The difference from Google and Microsoft’s approaches is not subtle: Amazon did not ask permission. It scraped product data from independent retailers, listed their products on its platform, and forced merchants who objected to opt out rather than opt in.
The backlash arrived in January. Bobo Design Studio CEO Angie Chua discovered she was part of the program only when orders began arriving from unfamiliar “buyforme.amazon” email addresses. Her viral Instagram post prompted more than 180 small businesses to reach out with similar complaints. Products that had been deleted from her backend were being sold; AI-generated images of items she had never carried appeared in listings.
The scale is significant. Amazon grew from 65,000 scraped products at launch to over 500,000 by November 2025, adding roughly 60,000 new products monthly. The company claims a zero-commission model, but the strategic logic is straightforward: expand its product catalog without inventory costs while making Amazon the default shopping interface.
The hypocrisy has not escaped notice. In November, Amazon sued Perplexity AI for letting its Comet browser make purchases on Amazon’s platform without authorization, alleging the startup disguised automated browsing as human activity. Perplexity responded by calling the lawsuit a “bully tactic” and noting that Amazon was doing the exact same thing to independent retailers. The company pointed out that AI agents shopping on Amazon’s behalf could not see the advertising Amazon typically shows users, which may be the more fundamental objection.
What the Retailers Want
Forrester analyst Sucharita Kodali has questioned whether chat-based commerce solves an actual problem, noting that it is unclear what value it brings retailers “other than disintermediating Google.” But for merchants, the question is simpler: which approach preserves their customer relationships?
The consent-based models offer obvious advantages. Under Google’s UCP and Microsoft’s Copilot Checkout, retailers remain the merchant of record, retain customer data, and control fulfillment. Shopify merchants are being automatically enrolled in Copilot Checkout but can opt out; the company says the opt-out process will be managed through Shopify’s admin interface.
Amazon’s extraction model offers no such protections. Fulfillment, returns, and customer service are handled by the brand, but Amazon controls the discovery interface and the customer relationship. For brands that have deliberately avoided Amazon’s marketplace for financial or strategic reasons, forced enrollment represents a significant loss of control.
The stakes extend beyond individual transactions. Mikey Vu, who heads the retail AI practice at Bain & Co., told Modern Retail that retailers remember how travel aggregators and flash-sale sites captured the top of the funnel in previous decades. Many are determined not to repeat the experience with AI agents.
The Advertising Question
Perplexity’s legal argument against Amazon identifies what may be the real issue at stake. In its November filing, the company argued that Amazon’s central concern is its inability to sell products to human users through ads if AI agents use the platform instead.
The point generalizes. If checkout happens inside AI assistants, the advertising-supported model of e-commerce becomes harder to sustain. AI agents do not see banner ads, do not browse related products, and do not respond to upsells. They execute transactions efficiently, which is precisely what retailers fear.