Preparing Your Store for AI Shopping Agents

Here’s the TLDR:

  • What this post covers: A quick prep guide for e-commerce brands to support AI agent transactions via Universal Commerce Protocol (UCP).
    • Core requirements for agent-compatible stores:
      • Product data: Unique consistent IDs, accurate prices (in cents), real-time inventory, image URLs
      • Discounts: Only percentage or fixed_amount types; no complex stacking or human-judgment rules
      • Shipping: Flat rates by country_code with default fallback; clear titles; free shipping via min_subtotal threshold
      • Checkout handoff: Cart state must persist through continue_url for transactions requiring human completion
    • Key platforms: Google AI Mode, Gemini app, Shopify Agentic Storefronts, ChatGPT, Microsoft Copilot
    • Merchant action required: Clean product feeds, simplify discount logic, configure shipping rates, verify Merchant Center health, submit UCP integration interest form
    • What breaks agent checkout: Inconsistent product IDs, price mismatches between feed and checkout, complex discount eligibility rules, real-time carrier rate shopping, inventory sync delays

There’s been a flurry of releases and announcements around agentic shopping lately.

Google, Shopify, and other major platforms are building systems that let AI agents complete purchases on behalf of your customers.

They’ve called it Universal Commerce Protocol, or UCP, for short.

An open standard that lets any AI agent transact with any merchant through a common set of rules.

I’ve dug into what information is available so far, and put together some general thoughts and a check list of sorts.

So you can begin preparing your ecommerce stores for this massive shift that’s on the way.

Instead of a human browsing your store, clicking through products, and checking out, an AI assistant could handle the entire transaction inside a chat interface.

The customer chats “I need flowers for my mom’s birthday, delivered tomorrow.”

The agent finds products, compares options, applies a discount code, selects shipping, and completes payment.

Your store receives an order.

The customer may never visit your website.

This is already rolling out.

Google’s AI Mode in Search and the Gemini app will support direct checkout from participating merchants.

Shopify is integrating this across ChatGPT, Microsoft Copilot, and Google’s surfaces through what they’re calling “Agentic Storefronts.”

If your store’s data is clean and your checkout logic is simple, agents can buy from you.

If it’s messy or complicated, they can’t.

Note: The below is based on limited information and initial implementation examples they’ve provided so far. This will certainly change over time, but the general sentiment around clean feed data will stand no matter the changes.

Your Product Data Is the Foundation

When an agent processes a transaction, it pulls product information from structured data feeds. Not by reading your website. The agent needs things like:

  • A unique product ID
  • A title
  • A price
  • An image URL
  • Inventory quantity
clean product data for agentic shopping

What makes this work:

  • IDs are consistent and descriptive (bouquet_roses, not SKU-39281-A)
  • Titles are human-readable (what an agent would show a customer)
  • Prices are accurate and match what your checkout will actually charge
  • Every product has an image URL

What breaks this:

  • IDs that change or differ between your feed and your actual database
  • Prices in your feed that don’t match prices at checkout
  • Missing images (agents often display product images in their interface)
  • Titles full of SEO keywords instead of plain descriptions

Inventory Must Be Accurate.

Agents will check inventory before completing a purchase. If your feed says you have stock but your actual inventory is zero, the transaction fails.

Action items for Product Data:

  • Audit your product feed for ID consistency
  • Verify prices match between your feed and your checkout
  • Ensure inventory syncs in near-real-time
  • Add image URLs to any products missing them

Discounts That Agents Can Use

Agents will apply discount codes the same way a customer would.

They send the code, your system validates it, and the discount gets applied to the total. This only works if your discounts are simple and programmatic.

Here’s what agent-compatible discounts could look like:

discount code data for agentic shopping

Two types: percentage (value is the percent) and fixed_amount (value is cents).

What works:

  • Percentage off entire order
  • Fixed dollar amount off
  • Free shipping over a minimum (covered in the shipping section)
  • Automatic discounts based on cart total

What doesn’t work:

  • “Buy 2 get 1 free” with complex eligibility rules
  • Discounts that require human judgment (“valid for first-time customers only” when the agent doesn’t know if this is a first-time customer)
  • Stacking rules that require interpretation
  • Time-limited codes where the validity logic is complicated
  • “Call for pricing” or “request a quote”

If your discount requires a person to evaluate whether it should apply, an agent can’t use it. The agent will either skip the discount, or the transaction will fail, or the customer will need to be handed off to your website to complete checkout manually.

Action items for Discounts:

  • Review your active discount codes
  • Identify any that require human judgment
  • Simplify or remove complex stacking rules
  • Make sure discount validation returns clear yes/no responses

Shipping Options Agents Can Understand

When an agent reaches the fulfillment step, it requests shipping options for the customer’s address. Your system returns a list of options with prices and descriptions. The agent presents these to the customer, they pick one, and the transaction continues.

Here’s how shipping rates could be structured:

shipping information for agentic shopping

The country_code field determines which rates apply. “default” is the fallback for any country not explicitly listed. So a US customer sees “Standard Shipping ($5)” and “Express Shipping (US) ($15)”. A customer in Canada sees “Standard Shipping ($5)” and “International Express ($25)”.
The title field is what the agent shows the customer. Make it clear.

Free shipping works through a promotion system that checks conditions:

  • Minimum subtotal threshold (“Free shipping on orders over $50”)
  • Specific eligible products (“Free shipping on all bouquets”)

When conditions are met, the standard shipping price becomes $0 and the title updates to “Standard Shipping (Free)”.

What works:

  • Flat rate by region
  • Free shipping over a threshold
  • Clear delivery timeframes in titles
  • Simple zone-based pricing

What doesn’t work:

  • Real-time carrier rate shopping during checkout (too slow, too variable)
  • Shipping rules that require product dimensions you don’t have in your feed
  • “Contact us for shipping quote”
  • Options that change based on factors the agent doesn’t have access to

Action items for Shipping:

  • Define shipping rates for your key regions
  • Set up “default” fallback rates for everywhere else
  • Write clear, descriptive titles for each option
  • Configure free shipping thresholds if you offer them

When Customers Land on Your Website Anyway

Not every transaction completes inside the agent. Sometimes the agent might hand it off to your website, and the customer finishes checkout there. This happens when:

  • Your checkout has requirements the agent can’t handle (custom product configuration, terms acceptance, identity verification)
  • Payment processing requires additional authentication
  • Something goes wrong and the agent can’t recover

When this happens, the agent provides a continue_url.

A link to your checkout with the cart already populated. The customer clicks through and picks up exactly where the agent left off.

This is where your website still matters.

What makes this seamless:

  • Cart state is preserved (items, quantities, selected shipping, applied discounts)
  • Customer doesn’t have to re-enter information
  • Your checkout works on mobile (most agent interactions happen on phones)
  • Clear visual confirmation that their cart is correct

What makes this painful:

  • Cart is empty or different when they arrive
  • They have to re-enter shipping address or payment info
  • Discount code that was applied in the agent is rejected on your site
  • Checkout is broken on mobile

If your checkout experience is frustrating after a handoff, you lose the sale.

The customer just spent time with an agent getting everything set up, and now they’re starting over.

They’ll leave.

Action items for Agent Handoff:

  • Test your checkout with pre-populated carts (Shopify handles most of this, but verify)
  • Ensure discount codes work the same way on your site as they do via API
  • Verify mobile checkout works smoothly
  • Consider the visual experience—the customer is arriving mid-transaction, not browsing

Merchant Center Becomes Your Control Panel for Google

For Google’s implementation, your Merchant Center account is where agent commerce gets configured. This is already used for Shopping ads and free listings, but it becomes even more important now.

Google is adding new data attributes specifically for “conversational commerce”.

Structured information that helps agents answer product questions and match products to customer needs. These go beyond traditional keywords to include things like:

  • Answers to common product questions
  • Compatible accessories
  • Substitute products
  • Use cases and occasions

These attributes are rolling out gradually. When they become available, using them properly will affect how often agents recommend your products.

Full Checklist: Preparing for Agentic Shopping

  • PRODUCT DATA
    • Product IDs are consistent between feed and database
    • Titles are clear, human-readable descriptions
    • Prices are accurate (in cents)
    • Every product has an image URL
    • Inventory counts sync in near-real-time
  • DISCOUNTS
    • Active codes are percentage or fixed_amount only
    • No codes require human judgment to validate
    • Stacking rules are simple or nonexistent
    • Validation returns clear yes/no
  • SHIPPING
    • Rates defined for key countries
    • “Default” fallback rates exist
    • Titles are clear and descriptive
    • Free shipping thresholds are configured
  • CHECKOUT HANDOFF
    • Cart state preserves through continue_url
    • Mobile checkout works smoothly
    • Discounts apply the same way via API and website
Written by Tony @ Unlocked Agency Google Ads, Shopify, AI Consultant