Agentic Commerce: Preparing Your Store for AI Shopping Agents

What is agentic commerce, and how much of it is live?

Agentic commerce is shopping where an AI agent does part of the work a person used to do: searching, comparing, filling a cart and, in some cases, paying. In November 2026, agents mostly find and compare; they check out only with a small, selected group of merchants.

It helps to separate three levels, because each asks something different of your store:

• Discovery. The agent recommends products and links to them. This runs on product feeds and your product pages, and it is open to almost every store.
• Cart and handoff. The agent builds a cart, sometimes across several stores, then passes the shopper to your checkout to finish.
• Native checkout. The shopper pays inside the AI product, and the order lands in your systems. This needs a technical integration and, today, approval from the platform.

Why bother now, when checkout is still a pilot? Because shoppers already arrive from AI tools. Adobe’s April 2026 analysis of US retail sites found that traffic from generative AI sources grew 393% year over year in the first quarter of 2026, and that in March this traffic converted 42% better than other traffic. Adobe also scored how much of each page type AI tools can read, and product pages came out lowest of the page types it measured, at 66% on average.

Two caveats: Adobe counts visits referred by AI tools, not purchases made by agents, and the visibility score comes from Adobe’s own checker. Still, the direction is clear.

Which platforms and protocols matter for a US store?

Three matter most for US merchants right now: Google, through AI Mode and the Gemini app; OpenAI, through ChatGPT; and Microsoft, through Copilot. Underneath sit two protocols: shared technical rules for how an agent talks to a store’s systems.

Google: UCP and Universal Cart

Google announced the Universal Commerce Protocol on January 11, 2026, as an open standard covering discovery, buying and post-purchase support. It was co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and Google says it works with other agent standards, including the Model Context Protocol.

On October 15, 2026, Google began rolling out Universal Cart in the US across Search and the Gemini app. Shoppers can save items from AI Mode, keep products from several retailers in one cart and get alerts when prices drop or items return to stock. They then pay with Google Pay at participating stores or move the items into the merchant’s own cart. Google says checkout is “powered by the Universal Commerce Protocol” and that merchants remain the seller of record. Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and “many Shopify merchants” were named at launch. For the consumer side of AI Mode, see our guide to Google AI Mode.

Getting your own checkout into this is not self-serve yet. Google’s UCP guide (updated October 19, 2026) says the Merchant Center integration hub is in early access. It is “gradually rolling out in the United States,” with Australia and Canada expected early next year. Checkout is open to select merchants who submit an interest form and complete the technical work. Only listings with the native_commerce(checkout_eligibility) attribute show the Buy now button.

OpenAI: ChatGPT shopping and ACP

The Agentic Commerce Protocol is the framework behind commerce in ChatGPT. OpenAI’s developer docs say building with it “is open to all,” and list three flows a merchant implements for Instant Checkout: product feeds, checkout, and payments.

The checkout story changed this year, and OpenAI’s own pages now send mixed signals. On March 24, 2026, OpenAI said the first version of Instant Checkout, its in-chat payment option, did not offer “the level of flexibility that we aspire to provide.” Merchants can now use their own checkout. Shopify purchases complete on the store in an in-app browser, and retailers can build ChatGPT apps, as Walmart has. ACP was expanded to carry product feeds and promotions, with Target, Best Buy and Wayfair among the named partners. Yet OpenAI’s commerce documentation still says Instant Checkout is “currently available to approved partners.” Our reading: plan for discovery plus your own checkout, and treat in-chat payment as the exception. We cover the wider ChatGPT picture in how to get your business mentioned in ChatGPT.

Microsoft: Copilot Checkout

Microsoft’s agentic commerce page, dated September 23, 2026, describes Copilot Checkout as live on Copilot.com and the Copilot mobile app, and as a pilot for select customers. It is limited to English-language merchants selling to US buyers in US dollars. Shopify merchants are enrolled automatically, with controls in the Shopify admin. PayPal and Stripe merchants opt in through their payment provider, and Microsoft says it takes no commission. The feed setup uses Microsoft Merchant Center with UCP feed attributes.

One pattern runs through all three: the merchant stays the merchant of record. The agent takes the order; the refund requests, chargebacks and customer emails still come to you.

What do AI shopping agents read from your store?

Structured product data first, then your policies, then your pages. An agent comparing three standing desks pulls price, stock, variants and returns from feeds and markup, then reads the product page to confirm and fill gaps.

OpenAI’s product feed reference shows how literal this is. A discovery feed needs nine fields per row: item_id, title, description, url, brand, seller_name, image_url, availability and price. Each row is “one purchasable item or variant.” A row with missing or unrecognized availability is rejected, and a sale price is ignored unless it is lower than the regular price and in the same currency. Product and image URLs must be publicly accessible. Return fields such as a return window in days are optional.

Google is adding fields built for conversation. Merchant Center’s conversational attributes include question_and_answer, document_link (PDF manuals and instructions), related_product, item_group_title, variant_option and popularity_rank. Google says adding them “won’t impact the approval status of your existing products,” and you don’t need to repeat details already in your description or product highlights.

In store audits, the problems that matter here are rarely exotic. We often find price and stock that load with JavaScript after the page renders, feeds that update overnight while site prices change during the day, and return policies that live only in a PDF or a pop-up. A person works around those gaps. An agent comparing your listing with four others usually moves on.

How do you get product data agent-ready?

Make one system the source of truth for every product fact, then make sure every feed and page reads from it. Most data errors we see come from a hand-edited second copy of the catalog.

Give every variant its own row and a stable ID

Take a home-goods store selling a table lamp in three finishes. That is three rows, each with its own item_id and a shared group ID, and each linking to a URL with that finish selected. Google’s UCP guide strongly recommends variant_option with item_group_id and item_group_title for multi-variant products (up to 30 variant options per item). It warns that incomplete variant data can suppress the checkout button. Keep IDs stable across re-imports; if your feed ID differs from your checkout’s, Google lets you map it with merchant_item_id.

Keep price and stock current, not just correct

Suppose that lamp goes on sale at 9 a.m. and your feed refreshes at 2 a.m. For 17 hours, every agent quoting your feed shows the old price, and an agent checkout risks a price mismatch. Google recommends moving catalog updates to its Merchant API to keep them in sync in near real time, or sending supplemental price and inventory updates between scheduled feeds. OpenAI’s spec makes a related point: date fields “do not schedule price or availability changes,” so you have to send the update when the sale starts and again when it ends.

Write descriptions that answer comparison questions

Agents get questions like “will this fit a 60-inch desk?” Put dimensions, materials, compatibility and care in plain text in the description and in specific attributes. Use question_and_answer for the questions your support team hears most.

Fill in identifiers

Use GTINs where they exist. OpenAI accepts 8-, 12-, 13- or 14-digit GTINs with a valid check digit, and asks you to leave the field empty rather than invent one. For your own brand without GTINs, submit brand plus manufacturer part number (MPN).

Which store policies do agents need to see?

Returns, shipping and customer support, stated as data and not only as prose. When an agent can buy for someone, the policy becomes part of the product.

Google’s UCP guide (updated October 29, 2026) calls a Merchant Center return policy a merchant-of-record requirement and says checkout displays it. Each policy must include the return cost, the return window and a link to the full policy. You also need customer support details (a service page URL, email or phone number), which become the “Contact Merchant” link on the order confirmation. Products with regulatory warnings, such as California’s Proposition 65, need a consumer notice that appears on the checkout screen.

Some products can’t use Google’s checkout at all. The guide excludes subscriptions and installment plans, personalized items, used, refurbished and final-sale goods, pre-orders, freight shipping, age-restricted products, services, rentals and virtual items. For those, the native_commerce(checkout_eligibility) value must be empty or FALSE. If most of your catalog falls into these groups, discovery is where your effort belongs.

Do product pages and structured data still matter if agents use feeds?

Yes. Feeds tell an agent what you sell; pages and markup are how it checks that claim and how it fills in what the feed leaves out. OpenAI requires product URLs that are publicly accessible, ideally with the variant pre-selected. And many AI shopping answers still link to the product page itself.

Three things to check on your product templates:

• Visible without JavaScript. Name, price, stock, variants and the returns summary should be in the HTML the server sends, not injected later. In our experience, you can’t count on every crawler or agent running your JavaScript.
• Crawler access. If your robots.txt or CDN blocks AI search crawlers, your pages can’t support the agent’s answer. Our guide to AI crawlers explains which bots to allow and which you can block without losing visibility.
• Product and return policy markup. Merchant listing markup on each product page, and a site-wide return policy on your Organization markup.

For a standard policy, Google’s documentation recommends MerchantReturnPolicy nested under Organization, placed on one page rather than repeated everywhere. A minimal version looks like this:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Home Goods",
  "url": "https://www.example.com/",
  "hasMerchantReturnPolicy": {
    "@type": "MerchantReturnPolicy",
    "applicableCountry": "US",
    "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow",
    "merchantReturnDays": 30,
    "returnMethod": "https://schema.org/ReturnByMail",
    "returnFees": "https://schema.org/FreeReturn",
    "merchantReturnLink": "https://www.example.com/returns/"
  }
}
</script>

Products with a different policy get their own MerchantReturnPolicy under the Offer. For complex or frequently changing rules, Google suggests setting returns in Merchant Center or Search Console instead. Our schema markup guide covers the other types worth adding, and Shopify owners will find theme-specific fixes in our Shopify SEO guide.

What changes for payments, fraud and security?

Less than you might fear on payments, more than you might expect on security. You keep your processor and your liability; the security work is about not blocking the agents you want.

Payments: your processor, your risk decisions

Under ACP, ChatGPT calls your checkout endpoints, but OpenAI says “the actual checkout state and payment processing occurs on the merchant’s systems.” You validate the order, calculate and charge sales tax, assess fraud risk and accept or decline. Card details travel as a one-time delegated payment with a maximum amount and an expiry, through your own payment service provider (PSP); Stripe’s Shared Payment Token was the first compatible implementation. Google’s UCP checkout uses Google Pay with payment and shipping details saved in Google Wallet. Microsoft lets a merchant use one PSP at a time as its Copilot Checkout partner.

Agent orders arrive through your normal order, fraud and tax systems. Talk to your payment provider before you apply anywhere; in most setups it does the integration work.

Security: don’t block the agents you signed up for

Google’s UCP guide has a page on network and firewall settings (updated October 30, 2026), and it reads like a list of what goes wrong. Google doesn’t automatically retry 403 or 429 responses, so one blocked request can end a checkout. Google shares a small number of outbound IP addresses, so ordinary per-IP rate limits can trip. And the browser checks that stop scrapers, such as CAPTCHAs, JavaScript challenges and fingerprinting, will also stop Google’s server-to-server calls.

Google’s advice is to exempt only the dedicated UCP, profile (/.well-known/ucp) and sign-in endpoints, and to verify requests properly. The options are mutual TLS, API credentials or signed requests checked against Google’s published keys. It warns that attackers “can spoof plain HTTP headers,” so a user-agent allowlist alone is not security.

What should you do first? A readiness checklist

Work from data outward: what agents read, then what they check, then what lets them transact. None of these steps depends on a platform choosing you for a pilot.

1. Audit price and stock agreementPick 20 best sellers and compare the price and availability on the page, in Merchant Center and at checkout. Any mismatch is your first fix.
2. Clean up variants and identifiersOne row per variant, a shared group ID, stable item IDs and real GTINs or brand plus MPN.
3. Speed up updatesSend price and stock changes when they happen, through the Merchant API or supplemental updates, not only in a nightly feed.
4. Publish one set of policiesReturn cost, window and link, shipping times and support contacts, the same in Merchant Center, on the page and in markup.
5. Enrich your top productsSpecific descriptions, Q&A pairs, manuals and related products for the items that earn the most.
6. Check your pages and bot rulesProduct facts in the server HTML, AI search crawlers allowed, and firewall rules ready to exempt verified agent endpoints.
7. Then decide on checkoutUse the framework below to choose whether to apply now or wait.

Should you apply for agent checkout now?

Our view, by situation:

• You run Shopify. Much of this is already switched on through Shopify Catalog and automatic enrollment. Your job is data quality and checking the settings in your admin, not integration.
• You run your own platform, sell standard physical goods in the US and have developer time. Submit Google’s interest form and talk to your PSP about ACP and Copilot. Early participants learn how agent orders behave while volumes are small.
• Most of your catalog is custom, made to order, subscription-based or age-restricted. Google’s checkout excludes these. Put the effort into discovery: feeds, Q&A and pages.
• Your feed and site disagree today. Wait. An agent checkout on top of inconsistent data turns into failed orders and support tickets.

If you aren’t sure where your pages stand, our free SEO + GEO report analyzes your website. It checks which AI crawlers can read it, content visible without JavaScript and structured data, and ranks each issue from critical to low. For stores that want help across feeds, pages and AI visibility, see our SEO + GEO services. Across the 500+ brands we’ve served, the stores that adapt fastest to new search surfaces are usually the ones whose product data was already in order.

Common questions

Sources

1. OpenAI (March 24, 2026): Powering product discovery in ChatGPT
2. OpenAI Commerce docs: Key concepts (Agentic Commerce Protocol)
3. OpenAI Commerce docs: Product feed reference
4. Google Blog (January 11, 2026): New tech and tools for retailers to succeed in an agentic shopping era
5. Google Blog (October 15, 2026): Universal Cart launch in the U.S.
6. Google for Developers: UCP integration hub in Merchant Center (updated October 19, 2026)
7. Google for Developers: Prepare your Merchant Center account for UCP (updated October 29, 2026)
8. Google for Developers: UCP network and WAF configuration (updated October 30, 2026)
9. Google Merchant Center Help: How to use conversational attributes
10. Google Search Central: Merchant return policy (MerchantReturnPolicy) structured data
11. Microsoft Advertising: Agentic commerce and Copilot Checkout
12. Adobe for Business (April 16, 2026): AI traffic grows but retail sites lag in AI search visibility