An agentic commerce playbook for fashion retailers in 2026

KEY TAKEAWAYS Summary by the editors

1. Agentic commerce means an AI assistant discovers, compares and in some cases buys products for a shopper, with the retailer still fulfilling the order.
2. In 2026 two competing open standards matter most: OpenAI's Agentic Commerce Protocol, built with Stripe, and Google's Universal Commerce Protocol, announced in January 2026.
3. For most fashion retailers the first job is not a new integration but clean, current product data: titles, variants, prices, availability and return terms.
4. Under the published Agentic Commerce Protocol, the merchant remains the merchant of record and keeps responsibility for fulfilment, returns and customer support.
5. Liability for disputes that fall short of fraud, such as an agent buying the wrong size, was still unsettled in 2026, so retailers should treat early pilots as limited and measured.

What is agentic commerce in fashion?

Agentic commerce is a model in which an AI assistant does part of the shopping work for a customer: finding products, comparing them, tracking prices and, with the customer's confirmation, completing the purchase. In fashion this means a shopper might ask an assistant for a waterproof jacket under a stated budget and receive a shortlist drawn from several retailers, with a checkout step inside the assistant itself. The retailer still holds the stock, ships the parcel and handles the return.

The idea is new enough that definitions vary, and the evidence of scale is thin. The announcements are real, but published volumes of agent-led fashion sales are not yet available. A sober plan therefore treats agentic commerce as a channel to prepare for, not a replacement for the existing website.

It helps to separate three layers. Discovery is the assistant recommending products, which already happens in search and chat tools. Checkout is the assistant completing the transaction, which is where the new protocols operate. Post-purchase is everything after the order: delivery, returns, exchanges and support. Most announcements concern the first two layers, and the third is where fashion's cost sits, so a playbook that ignores returns is incomplete.

Which protocols and platforms matter in 2026?

Two open standards dominate the discussion. OpenAI introduced Instant Checkout in ChatGPT on 29 September 2025, powered by the Agentic Commerce Protocol (ACP), which it built with Stripe. At launch it covered US users buying from US Etsy sellers, and OpenAI named Glossier, SKIMS, Spanx and Vuori among Shopify merchants expected to join later.

Google announced its Universal Commerce Protocol (UCP) at the NRF show in January 2026. FashionNetwork reported that Shopify, Etsy, Walmart and Wayfair co-developed it, that Stripe, Visa and Adyen are payment partners, and that Zalando is among the named retailers. In March 2026 trade press reported that Gap would let shoppers check out inside Google's Gemini app, calling it the first major fashion company to work directly with Google on this model.

The landscape is moving. CXM Today reported that OpenAI had earlier struck deals with Walmart and Etsy but backed away from plans for in-app checkout, which is a reminder that any single platform's roadmap can change within months.

Read also

What is agentic commerce? How AI agents will shop for fashion customers

What should a fashion retailer prepare first?

The common requirement across platforms is structured product data that a machine can trust. OpenAI's feed specification asks for an identifier, title, description, URL, brand, seller name, image, availability and price, with each variant on its own row and an absolute, public URL for every item. Placeholders such as "n/a" are not allowed. Google's route runs through Merchant Center, where FashionNetwork noted that sellers need to enrich their data to stay visible.

Fashion makes this harder than most categories because a single style can carry dozens of size and colour variants, each with its own stock position. An assistant that offers a size that sold out an hour ago creates a failed order and an annoyed customer. OpenAI's guidance for most merchants is to send the full feed once a day and push changes through an API during the day.

1. Audit product titles, descriptions and images for accuracy and plain wording.
2. Make sure every size and colour variant has its own record with live availability.
3. Add GTINs or other standard identifiers where they exist.
4. Publish return terms in structured form and on a public page.
5. Check that prices, sale prices and promotions match the website at all times.

Ownership is the other half of preparation. Product data usually sits with merchandising or product information teams, pricing with commercial teams, feeds with e-commerce and payments with finance. Agentic channels make errors in any of these visible to a customer quickly. Naming one accountable owner for the agent channel, with a short list of contributors, avoids the situation in which a wrong price in a feed is nobody's problem.

How should checkout, payments and returns be handled?

Under the ACP flow, ChatGPT sends order details to the merchant's backend, which accepts or declines the order. Payment is passed as a token to the merchant's existing processor. OpenAI states that merchants remain the merchant of record and keep control of fulfilment, returns and customer support. In practice that keeps customer data, fulfilment and support inside the retailer's own systems.

Returns deserve explicit design. OpenAI's feed specification includes fields for whether returns are accepted, the return window in days and a link to the policy. Retailers should map those fields to their real policy and check what consumer law in each selling country requires.

What are the risks and limits?

• Disintermediation. Checkout inside an assistant may shift traffic away from a retailer's own site and raises questions about who owns the customer relationship and data. Gap said it supplies product data to Gemini in advance to retain control.
• Unsettled liability. Worldpay reports that no liability shift exists for agentic transactions and that disputes short of fraud, such as the wrong item being bought, are still being negotiated between merchant, issuer, acquirer and platform.
• Limited reach. Early programmes were restricted by country and merchant type, and Gap said loyalty points could not be used in the first release.
• Ranking opacity. OpenAI says checkout items get no ranking preference, and that ranking weighs availability, price, quality and seller status, so there is no guaranteed placement.

There is also a strategic question of dependence. If a large share of discovery moved into a small number of assistants, those platforms would control presentation, ranking and fees. OpenAI says merchants pay a small fee on completed purchases through Instant Checkout. Retailers should model what that fee, plus any change in conversion and returns, does to margin before committing, and should keep their own site and direct relationships strong whatever happens in assistants.

Read also

Returns policies when the buyer is an AI agent: rules to write now

How should a retailer sequence the work?

A reasonable order is data first, policy second, pilot third. Fix the product feed and return terms, because that work also helps conventional search and marketplaces. Then decide which platform to approach based on where your customers already ask questions. Finally run a small pilot with a limited range, track order accuracy, return rates and support contacts against the website, and agree internal ownership between e-commerce, finance and customer service.

Measurement should be agreed before launch. Useful indicators include feed validation errors, orders rejected for stock or price mismatches, return rate by channel, support contacts per hundred orders and the share of agent orders from new customers. None of these require special tooling, and together they show whether the channel adds sales or merely moves them from the website at a lower margin.

Frequently asked questions

What is agentic commerce?

Agentic commerce is shopping in which an AI assistant finds, compares and sometimes buys products for a customer. The customer confirms key details, and the retailer still fulfils the order and handles returns.

Is ChatGPT Instant Checkout available to fashion brands?

OpenAI launched it on 29 September 2025 for US users buying from US Etsy sellers, naming several Shopify fashion brands as expected later. Merchants apply through OpenAI's process, and press reports in 2026 suggested OpenAI had scaled back in-app checkout plans, so check current status.

What is the Universal Commerce Protocol?

It is an open standard Google announced in January 2026 for AI agents and merchants to interact, from discovery to checkout. Shopify, Etsy, Walmart and Wayfair co-developed it, and retailers need enriched data in Merchant Center.

Who is responsible if an AI agent orders the wrong size?

There is no settled rule. Under the ACP specification, settlement, refunds and chargebacks stay with the merchant and its payment provider, but allocation for non-fraud disputes is still being negotiated across the payments chain.

GuideThe complete guide to AI in fashion e-commerce, marketing and retailRead the complete guide