The way people shop online is changing.
The agentic commerce era has started, and AI agents are increasingly shopping on behalf of humans, searching, comparing, and completing purchases with minimal human input.
On Black Friday 2025, they influenced $14.2 billion in online sales globally, with $3 billion from the US alone, according to Salesforce.
In this guide, we cover what agentic commerce is, why you need to act now, and the two types of agents that will interact with your store. We also walk you through the exact steps to get ready.
What is agentic commerce?
Agentic commerce is an online shopping model where AI agents handle the entire purchase journey on behalf of a shopper.
The shopper’s only input is describing what they want. The agent does the rest, handling everything from product discovery, comparison, and even checkout.
This is not a future concept. It is already happening.
For example, Microsoft’s Copilot Checkout and Perplexity’s Instant Buy already let shoppers complete a purchase without leaving the conversation.
And it is only going to get better.
As the technology matures, the expectation is for agents to operate with minimal human intervention and deliver results that more precisely match what the shopper asked for.
Why you should act now
Shopping is moving from human-browsed to agent-driven. Instead of a customer visiting your store, an AI agent will. Instead of a shopper comparing products, an algorithm will. The shopper still has the final say, but their role is shrinking.
By 2030, Bain & Company estimates the US agentic commerce market could reach $300B to $500B, making up 15 to 25% of all e-commerce. Consumer trust will take time to build, but AI agents will certainly touch most online shopping in some capacity within five years.
Global retailers are already making the shift, and the results are showing.
Merchants using Salesforce Agentforce grew sales by 13% over Cyber Week 2025, compared to 2% for those without. Microsoft reports that when shopping intent is present, journeys with Copilot are 194% more likely to result in a purchase.
The conclusion is simple.
Your future customers will increasingly be AI agents shopping on behalf of humans. Agent-ready stores get found and get purchased from. The ones that are not ready get skipped.
So adapt or get left behind.
Two types of AI agents
There are two types of AI agents, differing in how they find your products and complete a purchase: browser-based agents and protocol-based agents.
Browser-based agents
These agents come to your website. They navigate your pages, read your content, interact with your store, and complete purchases directly on your site. A human may still be in the loop, reviewing or approving before the final purchase. OpenAI’s Operator works this way.
Browser-based agents are only as good as your site lets them be. They read the underlying structure of your pages: the accessibility tree. If your HTML is not semantic, or buttons are not labelled, or forms are not structured correctly, the agent cannot navigate your store.
Note: WebMCP is an emerging browser standard by Google and Microsoft that gives AI agents explicit instructions on how to interact with your site, rather than having them figure it out. Currently in early preview on Chrome Canary.
Protocol-based agents
These agents never visit your website. Instead, when you submit your product catalog to Google Merchant Center or Microsoft Merchant Center, your products are indexed and become queryable by AI agents. They shop by querying those indexes directly.
The agent reads your product catalog (or feed), evaluates your product, and completes the purchase by calling your checkout API directly using the Universal Commerce Protocol (UCP) or the Agentic Commerce Protocol (ACP). The human reviews and approves before the order is placed.
The Protocol Stack Behind Agentic Commerce
For agents to shop on your behalf, they need a common language to talk to merchant stores. That is what commerce protocols do.
Three protocols define that language right now.
WebMCP
WebMCP is a browser standard proposed by Google and Microsoft. It handles how browser-based agents interact with your website. Instead of agents navigating your pages through guesswork, your site declares specific actions agents can perform: search, add to cart, checkout, etc. Agents know exactly what to do and how to do it.
Universal Commerce Protocol (UCP)
UCP is an open standard led by Google, co-developed with Shopify and Walmart. It handles how protocol-based agents discover your store, query your product catalog, and complete purchases without ever visiting your website. Google AI Mode and Microsoft Copilot both run on UCP. Because it is an open standard, any agent platform can adopt it.
Agentic Commerce Protocol (ACP)
ACP is OpenAI and Stripe’s protocol. It serves the same purpose as UCP but operates within the ChatGPT ecosystem. Unlike UCP, it is not an open standard. Merchants submit product feeds to an OpenAI-provided endpoint, and checkout happens inside ChatGPT. Stripe powers the checkout infrastructure, but merchants process payments through their own payment provider.
How to get your online store ready for agentic commerce
As covered above, there are two types of agents that will interact with your store, and each requires a different approach.
Here is what you need to have in place.
Step 1: Make sure agents can find your products
Protocol-based agents shop by querying product feeds and shopping indexes, not by browsing websites. If your products are not in those feeds, agents will not find them.
There are two platforms you need to be on: Google Merchant Center and Microsoft Merchant Center. Set up both, make sure your store settings are complete, and update your product feeds to support Universal Commerce Protocol (UCP).
UCP is the open standard that allows AI agents to discover, evaluate, and transact with your store. Both Google and Microsoft use this protocol, and a UCP-compliant feed is necessary for your products to be eligible for agentic checkout on either platform.
Shopify co-developed UCP alongside Google. If you are on Shopify, you are automatically enrolled on both platforms with no manual setup required, with the option to opt out.
Step 2: Make sure your product information is complete
Protocol-based agents
For protocol-based agents, the agent’s entire evaluation of your product happens inside your feed. If a field is missing, the agent either skips your product or gets it wrong.
Your feed needs to include:
| Field | Why it matters |
|---|---|
| Title and description | How agents identify and match your product to the shopper’s goal |
| Price and currency | Agents compare prices directly across stores |
| Availability | Out-of-stock products are skipped |
| Images | Required for product listing |
| Shipping details | Delivery promise agents evaluate directly |
| Return policy | Required for checkout eligibility |
native_commerce: TRUE | Opts your product into agentic checkout; without this, it is ineligible |
merchant_item_id | Maps your feed product ID to your checkout API |
Browser-based agents
For browser-based agents, structured data on your product pages is important.
Structured data is code (usually JSON-LD) added to your product pages that labels information in a way machines can reliably read. It tells the agent exactly what it needs to know about your product: the price, the currency, the return terms, and more.
Here are the key schema types you need on every product page:
| Schema type | What it tells the agent |
|---|---|
| Product + Offer | Name, price, currency, availability, and condition |
| AggregateRating + Review | Rating score and review count |
| OfferShippingDetails | Handling time, transit time, and delivery destination |
| MerchantReturnPolicy | Return days, fees, method, and refund type |
| inProductGroupWithID | Variant grouping – prevents agents from comparing your own SKUs against each other |
We have a detailed guide on structured data for AI agents that walks you through implementing it on your online store.
Step 3: Make sure agents can complete the purchase
Protocol-based agents
Protocol-based agents do not go through your storefront checkout. They complete the purchase by calling your checkout API directly using UCP.
To support this, you need to implement four API endpoints:
- Create session: the agent initiates a checkout with the items, buyer location, and fulfillment preferences.
- Get session: retrieves the current session state.
- Update session: handles shipping address changes, tax recalculation, and full checkout hydration before payment.
- Complete session: finalises the transaction. Your server receives a tokenised Google Pay payment, processes it through your payment provider, and returns the order ID.
- Cancel session: handles abandoned sessions.
Performance requirements are strict. Your create endpoint must respond in under one second. Your complete endpoint must respond in under six seconds. Google requires at least 95% uptime.
For Shopify merchants, this is handled automatically. For all other platforms, this is developer work. Full implementation guide: Google UCP Native Checkout.
Browser-based agents
These agents come to your website and interact with it directly. Three things determine how reliably they can do that.
Implement WebMCP. Declare exactly how agents should interact with your key flows: search, add to cart, and checkout. Instead of agents guessing from your HTML, your site tells them directly. Start with your search form; it is the lowest effort and highest impact.
Keep your accessibility tree solid. Agents use it for everything not covered by a declared WebMCP tool. Use semantic HTML, label all form fields, and avoid custom components that are not built with the correct underlying HTML.
Minimize Cumulative Layout Shift (CLS). Agents rely on element positioning to interact with your page. If elements shift after load, agents can click the wrong target or lose their place in a flow entirely.
Once navigation works, the checkout barriers to address are:
- Guest checkout by default: mandatory account creation will stop an agent immediately.
- No CAPTCHA in the purchase flow: most CAPTCHA systems treat agents as bots.
- JavaScript-independent checkout: SPA (Single Page Application) checkouts that rely entirely on JS to render are fragile for agents.
- Transparent pricing throughout: if shipping costs or taxes appear only at the final step, the agent cannot reconcile them against the user’s original request and will abandon the session.
Fix these, and a browser-based agent can move through your store the same way a human would. Just faster, and without hesitation.
Step 4: Test your store from an agent’s perspective
Run these checks before you consider your store agent-ready:
- Structured data: Use Google’s Rich Results Test to confirm your product schema is valid and complete. Errors here mean agents are working with incomplete information.
- Accessibility tree: Open Chrome DevTools, go to the Accessibility panel, and inspect your product and checkout pages. If the tree is shallow or unlabeled, a browser-based agent will struggle to navigate.
- Guest checkout: Go through your own checkout as a guest, with no saved credentials, on a fresh browser session. If you hit a registration wall, an agent will too.
- Product feed: Check your Google Merchant Center and Microsoft Merchant Center dashboards for feed errors and disapproved products. Disapproved products are invisible to agents.
- UCP attributes: Confirm
native_commerce: TRUEis set on products you want eligible for agentic checkout. Without it, they are excluded regardless of everything else.
These checks take less than an hour. Do them before you assume your store is agent-ready.
FAQs
No. Most of what is required builds on what you already have. Structured data is added to existing product pages. Product feeds are submitted through Google Merchant Center and Microsoft Merchant Center. WebMCP is implemented as a progressive enhancement on top of your existing site. The UCP checkout API requires development work, but it does not require a rebuild. Start with the highest impact steps first: your product feed and structured data.
Agents will skip you. Protocol-based agents query product feeds and indexes; if your products are not there, they are invisible. Browser-based agents will attempt to navigate your site, but if they hit a registration wall, a CAPTCHA, or an inaccessible checkout, they will abandon the session. Your products may still appear in agent recommendations, but the purchase will not complete. That sale goes to a competitor whose store works.
No, but large retailers will move faster. The infrastructure, UCP, ACP, and WebMCP, is open and available to any merchant regardless of size. Shopify merchants are already auto-enrolled on UCP through the platform’s partnership with Google. The steps in this guide apply to stores of any size. The merchants who act early, regardless of size, will have a head start.
AUTHOR
Melwyn holds a Bachelor's degree in Computer Science and has been writing professionally for over 4 years. He has a strong background in SEO and is certified in web accessibility from the World Wide Web Consortium (W3C). When he's not writing about website optimisation, you can find him enjoying a game of FIFA or exploring the open world of GTA.
Melwyn Joseph 
