Agentic Commerce: The Complete Guide for Brands and Retailers (2026)

1. 01

   A shopper describes the need in a full sentence and gets back three to five products. No results page, no ten open tabs.

2. 02

   The agent builds that shortlist by reading product data. Titles, attributes, prices, stock and reviews are its raw material, and a detail nobody wrote down simply doesn't exist for it.

3. 03

   In agentic commerce, your product data is your storefront. The agent never sees your homepage.

4. 04

   The rails already exist: OpenAI and Stripe's Agentic Commerce Protocol, Google's Universal Commerce Protocol, the Agent Payments Protocol and agent-ready card tokens from Visa and Mastercard.

5. 05

   The work that decides who gets picked is less glamorous: complete, consistent SKUs published wherever agents read. That practice is Agentic Commerce Optimization.

Take one purchase. A shopper types: "Waterproof trail running shoes for wide feet, good on mud, under $150, delivered by Friday."

1. 1

   The shopper states a goal

   One sentence holds a category, hard constraints (waterproof, wide fit, under $150), a use case (mud) and a deadline.

2. 2

   The agent fans out the request

   It splits the prompt into sub-queries: waterproof membranes, wide-fit lasts, lug depth for mud, a price cap, delivery options. Google calls this query fan-out in AI Mode.

3. 3

   It retrieves candidates

   From merchant feeds (Google Merchant Center, OpenAI product feeds, Shopify Catalog), crawled product pages and their structured data, marketplaces and review sites.

4. 4

   It filters on hard constraints

   Anything whose data doesn't confirm "waterproof", "wide" or a price under $150 is dropped. Stock and delivery data cut the list again.

5. 5

   It ranks and explains

   It keeps the products it can justify best and writes a one-line reason for each.

6. 6

   It hands over or checks out

   The shopper clicks through to the retailer, or confirms checkout inside the chat where the merchant supports it.

With good data, the answer looks like this.

ChatGPT

Illustrative demo

ChatGPT

ChatGPT

Every reason quoted comes from something the merchant wrote. A better shoe with a vaguer listing never shows up.

Where catalogs win or drop out

When we audit a catalog against this sequence, filtering is where most products quietly die. Retrieval problems are loud (a feed error, a disapproval). Filtering problems make no noise.

What the agent actually reads

Agents read text and structured fields, not your design. OpenAI's product feed specification spells this out: nine fields are required for every product (item_id, title, description, url, brand, seller_name, image_url, availability, price), and optional fields such as material, color, size, gender, age_group, dimensions, review_count and return_policy give the agent more to match on (OpenAI).

Keep item_id stable. Some exports regenerate IDs on re-import, and the feed then treats every product as new.

Google added conversational attributes to Merchant Center, including question_and_answer, related_product, variant_option and popularity_rank, built for the way people ask questions in AI Mode (Productsup).

For each surface in depth, read how to get your products recommended by ChatGPT and how products get picked in Google AI Mode and Gemini.

Why protocols matter to marketers

Protocols decide how a purchase happens. They have no say in which product gets recommended. ACP, for example, uses a Shared Payment Token scoped to one merchant and cart total, so the agent can pay without ever seeing the buyer's card (Stripe). Good for trust. Irrelevant to a product whose attributes fail the shopper's constraints.

UCP has more than 20 partner endorsements, including Adyen, American Express, Mastercard, Stripe, and Visa, and retailers such as Best Buy, Macy's, The Home Depot and Zalando (IT Brief).

A lot of agentic commerce advice starts with the protocols. For most brands that's backwards. Unless you run a custom stack, your platform, feed tool or payment provider will implement ACP or UCP for you. OpenAI's March 2026 refocus away from in-chat checkout shows how fast the plumbing changes. The catalog is the part that stays yours.

A protocol connects you to the agent. Your product data decides whether the agent picks you.

The full technical comparison of each agentic commerce protocol, from ACP to AP2, is in our protocols guide.

Agentic commerce use cases in 2026, and the data each needs

Count how many facts in the right column are structured fields on your pages. In catalogs we audit, it's usually a minority. Our ACO playbooks by industry go deeper, from fashion and luxury jewelry to beauty, home, and food.

What changes inside the organization

• Product data becomes a marketing asset. Catalog and marketing teams now share ownership of the fields agents read.
• Feeds matter as much as the website. Merchant Center and AI feeds are where many agents read first. On Shopify, a metafield shown on the PDP doesn't reach Merchant Center unless something maps it. See product feed optimization.
• Localization decides if you get recommended in each market. Agents answer in the shopper's language. A translated size chart still in inches on the Italian store is a classic mistake. See ecommerce localization.
• Scale is the real constraint. 20,000 SKUs in five languages is 100,000 product pages to keep complete.

If we could fix only one thing in a new catalog, it would be the first row below, starting with the top sellers. Titles get the attention. Attributes do the filtering.

The ten-point readiness checklist

• Complete core attributes

  Material, color, size, fit, dimensions, weight, compatibility. In structured fields, for every SKU

• Specific titles

  Product type first, then the key attribute and what sets it apart: "Women's Waterproof Trail Running Shoe, Wide Fit"

• Intent-rich descriptions

  Who the product is for, when to use it, what problem it solves, how it compares

• Answers to real questions

  FAQ and question_and_answer content covering sizing and care, compatibility, who it's not for

• Consistent data everywhere

  One price and one set of specs on the PDP, in Merchant Center, in AI feeds and on marketplaces

• Fresh price and stock

  Feed updates on every price and stock change. A weekly refresh is too slow

• Identifiers

  gtin, mpn and brand filled, with one GTIN per variant (never the parent's code on every size)

• Structured data on PDPs

  Product and Offer schema matching visible content, including shippingDetails and hasMerchantReturnPolicy (structured data guide)

• Every market and language

  Complete, localized data in each market you sell in, sizes and units included

• Category demand covered

  Category pages built around how shoppers phrase category questions (category page optimization)

Here's what fixing one product looks like, using the trail shoe from earlier.

Shopify agentic commerce: what Shopify merchants should do

Shopify made agentic storefronts available in its Winter '26 Edition in December 2025, so merchants can sell through ChatGPT, Perplexity, and Microsoft Copilot, with more channels coming. Shopify advises merchants to define structured product data, use attributes and metafields agents can read, and give agents store policies and FAQs to answer questions (BetaKit).

Shopify handles the connection; what the catalog says is on you. Assign each product a category from Shopify's standard product taxonomy and fill its category metafields, instead of burying size or material in the description HTML.

The ACO loop

Readiness isn't a one-off. Run it as a loop.

This loop is the core of Agentic Commerce Optimization. Agentic commerce is the change in how people shop; ACO is what brands and retailers do about it.