BECOME PART OF AVONAO

Choose your region/language

Select your desired region and language below.

Choose your language

The site for the European Union is available in six languages.

Choose your language

The global site is available in six languages.

Get in Touch

The Store Is Now an API: Retail Prepares for AI Shoppers

Avonao

When Adobe published its forecast for the U.S. holiday season in late September, one figure stood out: traffic from AI tools to retail websites is expected to rise 130 percent year over year, by 159 percent on Thanksgiving alone. Behind that number is an infrastructure that took shape in little more than a year – open protocols for checkout, product data written for machines and ways for stores to recognize the agents that visit them. The store increasingly has two audiences: people, and the software that shops on their behalf.

Adobe expects online spending of $275.1 billion between November 1 and December 31, 6.7 percent more than a year earlier. Traffic from AI tools, measured as visits from a link in an AI service, rose 693 percent during the 2025 holiday season. In July 2026, according to Adobe data reported by Digital Commerce 360, such visits converted 60 percent better and generated 53 percent more revenue per visit than other traffic. In Adobe’s survey of 5,000 consumers, 77 percent of those who had shopped with AI said they felt more confident in their purchase, and 69 percent said they were less likely to return it.

Salesforce, which measured that AI and agents drove 20 percent of retail sales worldwide during the 2025 holidays – about $262 billion – predicts that one in three e-commerce sites will offer their own shopping agent by this year’s Cyber Week. Morgan Stanley found in December that roughly 23 percent of Americans had bought something using AI in the previous month.

From answers to checkout

The step from AI that recommends to AI that buys came in stages. In September 2025, OpenAI introduced Instant Checkout in ChatGPT, together with the Agentic Commerce Protocol, an open standard developed with Stripe and published under the Apache 2.0 license. In November, Google began rolling out agentic checkout in Search in the U.S., with merchants such as Wayfair, Chewy and Quince. In January 2026, Google launched the Universal Commerce Protocol, UCP, developed together with Shopify, Etsy, Wayfair, Target and Walmart and endorsed by more than 20 companies, among them Adyen, American Express, Best Buy, Mastercard, Stripe, The Home Depot, Visa and Zalando.

In March 2026, OpenAI extended the Agentic Commerce Protocol to product discovery and let merchants use their own checkout. Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair integrated it for discovery, and Walmart launched an app inside ChatGPT with account linking and its own payments. In April, Amazon, Meta, Microsoft, Salesforce and Stripe joined the UCP Tech Council alongside its founders. At Google I/O in May, Google introduced Universal Cart, a single cart across retailers for Search, Gemini, YouTube and Gmail. Amazon’s own assistant, Rufus, was used by more than 300 million customers in 2025 and delivered nearly $12 billion in incremental annualized sales, according to the company.

The pace continued into the autumn. On September 8, Meta introduced Muse, a personal agent that can shop and asks the user before every purchase, with checkout through Stripe’s Link wallet. On September 24, Meta connected Walmart, Best Buy, Sephora, Wayfair and other retailers, along with Shop Pay and PayPal for payment. On October 6, Meta and Sierra announced the Personal Agent Protocol, an open standard for how personal agents interact with businesses, with Genesys, Shopify, Stripe and Walmart among its partners; its first specification is due later this month.

A catalog written for machines

For a merchant, the first requirement is to be found. An agent does not look at banners or page layouts; it reads structured product data. OpenAI accepts product feeds from approved partners as full daily snapshots of up to 500,000 items per file. Google added dozens of new conversational attributes to Merchant Center in January, and its Shopping Graph now holds more than 60 billion listings. Amazon’s Shop Direct lists more than 100 million products from more than 400,000 merchants that sell in their own online stores.

Platforms take on much of this work. Shopify Catalog makes merchants’ products available in ChatGPT, Perplexity and Microsoft Copilot; according to Shopify, millions of its merchants can sell to ChatGPT users, and every Shopify merchant is ready for UCP. PayPal’s store sync, Stripe’s Agentic Commerce Suite and Mastercard’s Agent Suite for Merchants offer comparable routes. In the first quarter, Adobe found that about a third of product pages could not be properly accessed by AI – the gap these services are designed to close.

Checkout as a protocol

Once a product is chosen, the agent needs a defined way to buy it. Under UCP, a business publishes a machine-readable manifest at a standard address on its website, /.well-known/ucp, that describes what it supports – cart, checkout, orders and identity linking – and agents connect through REST, MCP or A2A. The protocol separates the payment instrument from the payment handler and is compatible with Google’s Agent Payments Protocol, AP2. The Agentic Commerce Protocol takes a similar approach with OpenAPI specifications for checkout and delegated payments; its latest stable version, from April 2026, added cart, feeds, orders, authentication and MCP.

Payment credentials are limited to the purchase at hand. Stripe’s Shared Payment Token is valid for one merchant and one cart total; at stores outside Link’s network, Muse pays with a one-time virtual card limited to the approved purchase. AP2 records a user’s instructions in cryptographically signed mandates: an Intent Mandate for the request and a Cart Mandate that locks the exact items and price. Google is donating AP2 to the FIDO Alliance, which is also receiving Mastercard’s Verifiable Intent, a tamper-resistant record of what a user authorized.

Across these systems, one principle is consistent: the retailer remains the merchant of record. OpenAI, Google, Shopify, Microsoft, PayPal and the UCP specification all state it. The sale, and with it the customer relationship, stays with the store; the agent carries the conversation and hands over the transaction.

Knowing which agent is at the door

The third building block is identity. When an agent visits a store, the merchant needs to know whether it is a known agent acting for a real customer. Cloudflare introduced signed agents in August 2025, based on Web Bot Auth, in which agents cryptographically sign their HTTP requests; the first cohort included ChatGPT agent, Goose, Browserbase and Anchor Browser. Since September 1, the underlying specification, HTTP Message Signatures for automated traffic, has been a working-group draft at the IETF, with authors from Cloudflare and Google.

The card networks build on the same foundation. Visa’s Trusted Agent Protocol, developed with Cloudflare, uses HTTP Message Signatures to convey an agent’s intent, the recognition of the consumer and, optionally, payment information. In June, Visa added an Agentic Directory of verified agents and merchants and an Agent Score that rates how ready a merchant’s site is for agents. Mastercard registers the agents that use Agent Pay, and on September 30 it added a service, in testing in the U.S. with Cloudflare and Skyfire, that scores the probability that a transaction was initiated by an AI agent. American Express offers Agent Registration in its developer kit, together with purchase protection for purchases made by registered agents.

Trust, step by step

Consumers are adopting the technology in stages. In a Harris Poll conducted for Visa in May, 23 percent of U.S. consumers said they trust generative AI to handle payments on their behalf, and 61 percent said they would trust Visa to handle agentic transactions. In an earlier Visa study, 38 percent of U.S. adults said they were comfortable with AI completing a purchase, and 60 percent said they would not let AI spend without their approval.

The products reflect that preference. Muse asks before every purchase, and Visa Intelligent Commerce works with spending limits set by the consumer. Shopify’s guidelines for automated agents on its stores, in place since 2025, call for a final review by the buyer before an agent completes a payment and point developers to its Checkout Kit. In September, six banks – ASB, Bank of America, Capital One, CommBank, ING and NatWest – published shared principles for trusted agentic commerce, covering transparency, safety, privacy and data, choice and interoperability.

What the forecasts measure

Estimates of the market’s size depend on their definitions. Morgan Stanley expects agents to account for $190 billion to $385 billion of U.S. e-commerce by 2030, a share of 10 to 20 percent. Bain, which counts purchases that agents initiate, influence or complete, arrives at $300 billion to $500 billion. McKinsey estimates that agents could orchestrate $900 billion to $1 trillion of U.S. retail revenue by 2030, and $3 trillion to $5 trillion worldwide. For business purchasing, Gartner expects 90 percent of B2B buying to be intermediated by AI agents by 2028, with more than $15 trillion passing through agent exchanges.

Merchants see the shift in their own data. Shopify said in August that AI-driven traffic and orders to its merchants’ stores had tripled within a year. Microsoft reported that shopping journeys with Copilot were 194 percent more likely to end in a purchase when shopping intent was present.

The weeks before the season

Several milestones fall into the weeks before the holidays. The first version of the Personal Agent Protocol specification is due in October. OpenAI, whose ChatGPT now reaches 1.2 billion people each week, begins testing a new visual ad format in the U.S. later this month. And Mastercard projects that by 2030, one in ten consumers will routinely use agents to make purchases.

For retailers, the practical agenda follows from the protocols themselves: product data that machines can read, a checkout that agents can call and a way to recognize the agents that arrive. The store that people visit remains. Alongside it, a second storefront is taking shape – one that is read rather than seen.

Sources and further reading

14

Oct ’26
Enterprise AI

The Digital Colleague: How AI Agents Join the Team

In a matter of weeks this autumn, Salesforce, Microsoft, Cisco and Google each presented AI agents designed to work as teammates – some with an email address and an entry in the company directory. Behind the vocabulary is a practical shift: companies are building the identity, oversight and measurement that any new colleague requires.

Read more

06

Oct ’26
Mid-Market

More Work Than Workers: The U.S. Mid-Market’s AI Turn

America’s mid-market grew its revenue by 11 percent while hiring slowed to its post-pandemic low. The companies using AI are the ones still adding staff – and as millions of businesses change hands by 2035, AI is becoming part of what a company is worth.

Read more
Tags AI AI Agents Agentic Commerce Artificial Intelligence E-Commerce Retail Holiday Season UCP Payments
Share this article:

Avonao

 {07} NEWSLETTER

Stay in the
Avonao loop.

Product news, insights and offers, delivered straight to your inbox.

    By clicking “Subscribe,” you agree that we may send you our newsletter by email and track whether you open it and which links you click. You can unsubscribe at any time with the link in every email. More in our Privacy Policy.

    © 2026 Avonao. All rights reserved.

    Protected by Cloudflare

    OUR PLATFORMS

    LEGAL

    We're on social media