Agentic Checkout Protocols Explained for Marketers

Agentic checkout protocols let an AI assistant complete a purchase with your store without the shopper visiting your site. The two main ones are OpenAI's Agentic Commerce Protocol (ACP) and Google's Universal Commerce Protocol (UCP). In both, you stay the seller and keep payments on your own rails. For marketers, the work is mostly upstream: clean product data, clear policies and accurate offers, because those are what the agent reads.

Method: we read OpenAI's commerce documentation, Google's January 2026 UCP announcement and Shopify's help centre on 31 October 2026 (CET). Quotes are verbatim. These products change fast, so check the linked pages. Opinions are marked "Our view".

On this page

• Why should marketers care about checkout protocols?
• How does OpenAI's Agentic Commerce Protocol work?
• How does Google's Universal Commerce Protocol work?
• How do ACP and UCP compare?
• Where does Shopify fit?
• What do marketers actually own here?
• Checklist
• From MentionX data: how often shopping-style questions name brands
• Worked example: Ledgerline's ecommerce partner gets ready
• What data do agents need from you?
• How should marketing and engineering split the work?
• What should you watch after launch?
• Are there risks?
• Should every merchant join now?
• How do you keep up with changes?
• How do agent purchases affect promotions?
• What should customer service know?
• Common questions
• Sources

Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.

Why should marketers care about checkout protocols?

Because the path from question to purchase can now stay inside an AI answer. If an assistant can buy from you directly, the product page, the landing page and the cart may never load. The information the agent pulls from your feed and policies is your shop window.

Our view: you don't need to understand the API calls. You do need to know what data the agent uses, who owns the customer, and what gets reported back.

How does OpenAI's Agentic Commerce Protocol work?

OpenAI says supporting Instant Checkout in ChatGPT "requires a merchant to implement three flows" (OpenAI, Key concepts):

The checkout runs on your systems. OpenAI says: "The checkout session is rendered in the OpenAI UI, but the actual checkout state and payment processing occurs on the merchant's systems" (OpenAI, Key concepts).

The spec's own summary is "Enable merchants to run end-to-end checkout flows inside ChatGPT while keeping orders, payments, and compliance on their existing commerce stack" (OpenAI, Agentic Checkout Spec).

Who is the seller?

You are. "OpenAI is not the merchant of record in the Agentic Commerce Protocol" (OpenAI, Key concepts). The payment spec adds that "Settlement, refunds, chargebacks, and compliance remain with the merchant and their PSP" (OpenAI, Delegated Payment Spec).

How are payments kept safe?

The payment token is "restricted by the delegated payment's max amount and expiry" (OpenAI, Delegated Payment Spec). OpenAI names "Stripe's Shared Payment Token" as "the first Delegated Payment Spec-compatible implementation" (OpenAI, Key concepts).

Can anyone join?

Not yet. OpenAI says onboarding product feeds "is currently available to approved partners" (OpenAI, Get started). Its prohibited list includes "age-restricted products (for example, alcohol, nicotine, gambling)", "prescription-only medications" and "unlicensed financial products" (OpenAI, Get started).

How does Google's Universal Commerce Protocol work?

Google announced UCP on 11 January 2026 as "a new open standard for agentic commerce that works across the entire shopping journey — from discovery and buying to post-purchase support" (Google).

Key points from the announcement:

• UCP "establishes a common language for agents and systems to operate together across consumer surfaces, businesses and payment providers."
• It "is compatible with existing industry protocols like Agent2Agent (A2A), Agent Payments Protocol (AP2) and Model Context Protocol (MCP)."
• It was "co-developed with industry leaders including Shopify, Etsy, Wayfair, Target and Walmart."
• It "will soon power a new checkout feature on eligible Google product listings in AI Mode in Search and the Gemini app", starting with "eligible U.S. retailers".
• "Retailers remain the seller of record."

All quotes from (Google). The same post launched Business Agent, which lets shoppers "chat with brands, right on Search", and said eligible US retailers "can choose to activate and customize this branded agent in Merchant Center".

How do ACP and UCP compare?

Sources: (OpenAI; Google).

Where does Shopify fit?

Shopify connects stores to these channels. It says agentic storefronts let "customers discover and purchase your products in AI channels, such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta", and that the feature "is active by default for eligible stores" (Shopify).

Shopify describes the channels differently. On the page we read, "ChatGPT acts as a discovery-focused referrer platform", with users completing purchases "on your online store checkout in a ChatGPT in-app browser". For Google AI Mode and Gemini, Microsoft Copilot and Meta surfaces, customers can buy in "Shopify-powered direct checkout" without leaving the channel, if direct checkout is activated (Shopify).

Our view: that difference matters. Which checkout a shopper gets depends on the channel, your platform and your settings, so check your own admin before assuming.

Shopify also says: "you retain full ownership of the customer relationship and post-purchase experience", and orders show "channel or referrer attribution" (Shopify).

What do marketers actually own here?

Our view:

1. Product data quality. Titles, descriptions, images, price and availability in your feeds. OpenAI says recommended attributes "improve ranking, relevance, and user trust" (OpenAI). See Merchant Center and product feeds.
2. Policies an agent can read. Shipping, returns and warranty in plain text. Shopify's knowledge base holds "your return process, shipping timelines, and brand values, so that AI agents can answer customer questions" (Shopify).
3. Offers and promotions. Keep them consistent across feeds and site.
4. Attribution. Make sure reports separate AI-channel orders.
5. Eligibility. Check the prohibited-products lists before planning a campaign.

Checklist

• Ask your platform which AI channels you're enabled for, and which checkout each uses.
• Audit feed completeness against each channel's spec.
• Publish shipping and returns in plain, quotable language.
• Add AI channels to your sales reporting.
• Read how AI agents browse, compare and buy.

From MentionX data: how often shopping-style questions name brands

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

We don't track agent checkouts. Our answer data does show that comparison and recommendation questions, the ones that lead to purchases, name brands most often:

Engines also differ in how often they name brands at all:

Our view: agentic checkout sits at the end of the same journey. If your product isn't named in the recommendation step, a checkout integration won't help. Work on being recommended and on checkout readiness together.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

Worked example: Ledgerline's ecommerce partner gets ready

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline isn't a shop, so this example uses a fictional partner, a stationery retailer that sells through Ledgerline's invoicing. The retailer's platform offers an option to let AI agents check out. Before switching it on, the marketing team audits its product feed: some products lack GTINs, several prices differ from the site, and the returns policy is only in a PDF.

The team fixes identifiers, syncs prices from one source, publishes the returns policy as a web page and adds shipping times per region. Then it enables the platform option for a small range and checks orders weekly for errors.

What data do agents need from you?

Whatever the protocol, an agent completing a purchase needs reliable facts:

• Product identity: names, brands, GTINs or other identifiers, variants such as size and colour.
• Price and availability: current, in the right currency, with tax and shipping rules.
• Policies: returns, refunds, warranties and delivery times, in plain text on a web page.
• Images and descriptions: accurate and specific, so the agent picks the right variant.

If any of these are wrong, the agent may buy the wrong thing or give up.

How should marketing and engineering split the work?

Engineering or your platform provider handles the integration, payment and security. Marketing owns what the agent reads and says: product copy, offers, policies and feed quality. Agree who checks orders placed by agents and how errors are reported. Our view: name one owner for feed accuracy; most problems come from feeds nobody owns.

What should you watch after launch?

Track agent-placed orders separately if your platform allows it. Watch for wrong variants, price disputes, returns and failed checkouts. Compare return rates with other channels. If problems cluster on certain products, check their data first.

Are there risks?

Yes. An agent might present your product with an outdated price, or a customer might dispute an order they didn't fully review. Read the platform's terms on liability and disputes. Make sure your customer service team knows agent orders exist and how to spot them.

Should every merchant join now?

Not necessarily. If your feed and policies are already clean and your platform supports it with little work, a small pilot is low risk. If your data is messy, fix it first. The same fixes help you in shopping results and AI recommendations, so the work pays off either way.

How do you keep up with changes?

Protocols and platform features are changing fast. Subscribe to your platform's and payment provider's developer updates, and check the official documentation before each decision. Don't rely on summaries, including this one, for technical details; read the source and note the date you checked it.

How do agent purchases affect promotions?

Agents read the price and offer they find. If a discount code is only shown in an email or a pop-up, the agent may miss it. If a sale price expires but your feed isn't updated, the agent may quote the old price. Put active offers in structured product data and on the product page, with clear start and end dates, and remove them on time. Check how your platform handles coupon codes in agent checkouts before running a promotion.

What should customer service know?

Brief your support team on agent orders: how to recognise them, what the customer saw, and how to handle a dispute. Customers may say "the assistant bought it" and expect you to know what that means. A short internal note and a tag on these orders make support faster. Review the note whenever your platform adds a new agent feature.

Common questions

Do I need to build a protocol integration myself? Usually not. Platforms and payment providers offer integrations; check what yours supports.

What matters most for marketers? Accurate product data, prices and policies that agents can read.

Is agentic checkout widely used yet? It's early. Start with clean data and a small pilot.

Sources

All accessed 31 Oct 2026.

• OpenAI, Agentic Commerce key concepts; Agentic Checkout Spec; Delegated Payment Spec; Get started
• Google, New tech and tools for retailers to succeed in an agentic shopping era (11 Jan 2026)
• Shopify, Shopify agentic storefronts