How to prepare your store for AI shopping agents

In short: for an AI agent (ChatGPT, Gemini, Perplexity, Copilot, or whichever one your customer uses) to recommend and sell your products, your store has to be readable by a program. That means a catalog with complete attributes, structured product data on every product page, a feed that updates when price or inventory changes, shipping and return policies written without ambiguity, and a checkout that doesn't break when the purchase starts somewhere else. Most of that can be solved with what you already have in Shopify, WooCommerce or BigCommerce. What no app solves is having reliable inventory and prices, and that is where most stores fail. In the US the stakes are concrete: agentic checkout through Google is already live, and it depends on a clean Merchant Center account.

This guide is in priority order: first what makes you visible today, then what prepares you for agentic purchases, and at the end an agentic commerce readiness checklist you can run through in one afternoon.

Why you need to prepare your store

A shopping agent doesn't look at your store the way a person does. It doesn't see the banner or the design: it reads data. When a user asks for "a waterproof backpack for a 15-inch laptop, delivered by Friday, under 80 dollars," the agent looks for products that have those attributes declared, compares total price and delivery time, and discards whatever it can't verify.

That has an uncomfortable consequence: a store with good products and weak data loses to a store with similar products and complete data. The agent won't guess that your backpack is waterproof if that only appears in a photo.

What agentic commerce is and where it stands as of October 2026 is covered in what is agentic commerce. This article is the practical part. If you want to prepare your store for AI shopping agents, the work breaks down into catalog, structured data, feeds, inventory, policies and checkout.

The three layers an agent reads

The first two layers pay off today in any market. The third matters when an assistant lets people buy without leaving the chat, which as of October 2026 is already live at scale in the United States. That is why the third layer is no longer something to watch from a distance if you sell here.

Step 1: clean up the catalog

Everything else depends on this. A perfect feed built on poor data is still poor.

Titles that say what the product is

The title should name the product the way someone would search for it: product type, brand, model and the attribute that sets it apart.

Complete attributes, not long descriptions

The description is for people. Attributes are for agents. For each category, define which attributes are mandatory and load them as fields, not only in the text:

• Apparel and footwear: size (with a size chart), color, material, fit, gender.
• Electronics: model, compatibility, voltage, warranty, what's in the box.
• Home: dimensions, material, weight, capacity, care instructions.
• Beauty: skin type, main ingredients, volume, expiration date.
• Food and supplements: net weight, flavor, allergens, servings per container.

Identifiers

If your products have a barcode (GTIN, UPC or EAN), load it. It's the simplest way for an agent to know your product is the same one another store sells, so it can compare price and shipping. If you make your own products and they have no code, use a stable SKU of your own and don't change it.

Well-built variants

Every size and color combination should be a variant with its own inventory and its own price, grouped under the same product. A "t-shirt" product with sizes written in the description is invisible to an agent searching for "size L in stock."

Images

At least one image on a clean background showing the whole product, plus detail shots. Agents use images to display results, and some models read them to confirm attributes.

Step 2: structured data on every product page (product schema for AI)

Structured data is a block of code, invisible to the shopper, that describes the product in a standard format (schema.org) that search engines and agents read. The type that matters is Product, with an Offer inside it. Getting product schema for AI right is mostly about completeness: the fields below are the minimum.

On Shopify, the official themes already generate a good part of this. What is usually missing is shipping, returns and the barcode. There are apps that fill in the block, and you can adjust it in the theme.

On WooCommerce, the most widely used SEO plugins generate it, with varying levels of detail.

On BigCommerce, the default themes output basic product data, and you can extend it through the theme or an app. Check shipping and return fields in particular.

On a custom ecommerce build, it has to be generated in code. It's days of work, not weeks, if the data is already in the database.

How to verify it: Google's Rich Results Test tells you what it reads from a page and what it's missing. Test one product page per category, not just the home page.

Step 3: feeds that update on their own

A feed is the full list of your products in a file that refreshes automatically. It's what assistants and search engines read when they compare without visiting your store.

The feeds worth having

What a ChatGPT shopping product feed needs

OpenAI's product feed specification requires, at minimum, per product: item ID, title, description, URL, brand, seller name, image URL, availability and price. If you declare variants, it also asks how they are grouped. It accepts JSONL, CSV and TSV files. And it says to update price and availability when they change, not once a week.

One point about checkout. Buying inside ChatGPT requires a separately enabled integration. Since March 2026, OpenAI has focused on product discovery, and merchants use their own checkout to complete the sale. In practice, that makes your own checkout page, and how well it handles a buyer arriving from an assistant, the part you control.

Google Merchant Center and agentic checkout

Google is where the US reader has the most to do right now. Google's Universal Commerce Protocol (announced in January 2026) and agentic checkout in AI Mode and Gemini depend on your Merchant Center account. As of October 2026, this is available to US merchants, and the Universal Cart has been live in US Search and the Gemini app since October 15, 2026.

For Google Merchant Center agentic checkout, Google requires:

1. An active Merchant Center account.
2. Approved products, with no open disapprovals for price or availability mismatches.
3. Return policies declared in the account.
4. A per-product marker in the feed that says which products are eligible for checkout.

That last item is the one stores miss. Having products approved is not the same as having them marked as eligible. Open the Merchant Center diagnostics, clear the disapprovals, declare your return policy, and then check which products carry the checkout eligibility attribute. If you use an app or a platform integration to send the feed, confirm whether it supports that attribute before assuming it does.

Shopify's channel for AI assistants

Shopify has offered, since 2026, a sales channel that publishes the catalog to AI assistants such as ChatGPT, Copilot and Perplexity. Which assistants are available, and whether they allow payment inside the chat, varies by assistant. Check the sales channels section of your admin to see what is on offer for your store today.

The consistency rule

Price, inventory and shipping must be identical on the product page, in the feed and at checkout. Agents and search engines compare those sources, and a mismatch (a lower price in the feed than in the store, a product in stock in the feed and sold out at checkout) is grounds for disapproval in Merchant Center and a reason for an agent to distrust you.

Step 4: reliable inventory and price

This is the step no app solves on its own.

If you sell in more than one place (your online store, retail locations, Amazon, Walmart Marketplace, wholesale accounts), inventory has to come from a single source and reach every channel within minutes. Three typical situations:

Promotions deserve special attention. If you run sales often, a feed with yesterday's price is a feed with the wrong price, and an agent that quotes it to a shopper creates a dispute you have to resolve. Connecting your ERP or inventory system to your store is a project in its own right, with costs we detail in how much software integration costs.

Step 5: policies a program can understand

An agent buying on someone else's behalf evaluates risk. Clear policies lower that risk.

Write and publish, on a dedicated page and in the structured data:

• Shipping: zones, costs by zone or by weight, delivery times in business days, free shipping threshold.
• Returns and exchanges: window in days, conditions, who pays shipping, how to start one, when the refund is issued.
• Warranty: term and what it covers.
• Payment methods: which cards, wallets and buy now, pay later options you accept, and any surcharge.
• Company information: legal name, tax ID, address and a real contact.

Avoid anything that forces interpretation: "see terms," "ships nationwide, contact us for rates," "exchanges subject to availability." For a person that is annoying. For an agent it is a data point that doesn't exist.

Step 6: checkout and APIs

This layer matters when the assistant lets the buyer complete the purchase or build the cart from outside your store.

What you can do today on any platform

• Prebuilt cart links. Shopify, WooCommerce and BigCommerce all let you generate a link that opens checkout with products already loaded. It's what an agent uses to leave the buyer one click away.
• Guest checkout. Every extra step is an abandonment, wherever the buyer comes from.
• Familiar payment methods. Cards and wallets the user already has saved, such as Shop Pay, PayPal, Apple Pay and Google Pay, reduce friction when the buyer arrives from an assistant. If you use Stripe as your processor, check which wallets are turned on.
• Don't block legitimate agents. Review your robots.txt file and the rules in your firewall or fraud tool. Some configurations block assistant crawlers and automated purchases without distinguishing between a malicious bot and an agent bringing you a customer.

What depends on the platform

Agentic purchase protocols (Google's Universal Commerce Protocol and the Agentic Commerce Protocol from OpenAI and Stripe) are integrated on the platform or payment processor side. On Shopify they arrive as channels you turn on from the admin. On other platforms it is better to follow their announcements than to get ahead of them with custom development, unless you are on a custom build.

What to do if your store is custom

If your ecommerce is custom, preparation means exposing the following through an API, with documentation and authentication:

1. Catalog with attributes, price and inventory.
2. Cart creation and payment link.
3. Order status.
4. Return initiation.

With that, integrating a protocol is one more layer, and it is also what your own customer service agent needs. If you are still choosing a platform, the practical rule is that hosted platforms give you the technical layer almost for free, while a custom build gives you more control over what you expose.

Step 7: reviews and trust signals

Agents compare reputation. What helps:

• Verified reviews on the product page, with the count and average in the structured data.
• Responses to negative reviews.
• Consistent presence: the same name, address and contact on your store, on Google and on your social accounts.
• Content that answers real questions: size guides, comparisons between your products, frequently asked questions by category.

Asking for reviews at the right moment is one of the jobs a post-sale agent of your own can do. We explain it in AI agents for ecommerce.

Shopify, WooCommerce, BigCommerce or custom: what you have and what you're missing

What the table shows: on platforms, the technical side comes almost solved, and the work is in the data and the synchronization. On a custom build there is more technical work but also more control over what you expose.

Agentic commerce readiness checklist: review your store in one afternoon

Catalog

• Titles say product type, brand, model and key attribute.
• Each category has mandatory attributes loaded as fields.
• Products have a barcode or a stable SKU.
• Variants have their own inventory and price.
• There is at least one clean-background image per product.

Structured data

• One product page per category passes Google's Rich Results Test without errors.
• It includes price, currency, availability, brand and barcode.
• It includes shipping cost and return policy.
• It includes reviews, if you have them.

Feeds

• The catalog is connected to Google Merchant Center with no products disapproved for price or availability mismatches.
• Your return policy is declared in Merchant Center.
• You checked which products carry the checkout eligibility attribute in the feed.
• The Meta catalog is connected.
• Feeds refresh on their own when price or inventory changes.
• You reviewed which AI assistant channels your platform offers.

Inventory and price

• Inventory comes from a single source for every channel, including Amazon, Walmart Marketplace and retail locations if you use them.
• The price on the product page, the feed and checkout is the same.
• Promotions have a start and end date in the system and are not entered by hand.

Policies

• Shipping with zones, costs and delivery times published.
• Returns and exchanges with window, conditions and who pays.
• Company information and a real contact are visible.

Checkout

• A shopper can buy without creating an account.
• You can generate a link to a prebuilt cart.
• Shop Pay, PayPal or the wallets you want are turned on in your processor.
• Your robots.txt and fraud tool don't block the AI assistant crawlers you want reading your store.

If you check everything except the inventory and price items, you are ahead of most stores. If inventory and price are what fails, start there.

What it costs to get ready

Market reference ranges as of October 2026, for a small or mid-size business:

If you also want an agent of your own that uses that same data to answer questions and sell on WhatsApp and Instagram, at Atomix that project starts at USD 3,000, plus monthly maintenance from USD 300 and the costs of the AI model and the channel. How we build them is on our page about AI agents for business. What that agent does day to day is in AI agents for ecommerce.

Frequently asked questions

How do I get ChatGPT to recommend my store?

There is no way to pay for a place in organic answers. What improves your chances is having product pages with complete structured data, a catalog sent through the channels the assistant reads (directly or through your platform), reviews, and content that answers real questions. Consistency of price and inventory across sources also counts.

What is a product feed and why does it matter for AI?

It is a file with all of your products, their attributes, price and inventory, that updates automatically. Assistants and search engines use it to compare stores without visiting each site. Without a feed, they depend on crawling your pages, which is slower and less accurate.

What does a ChatGPT shopping product feed need to include?

At a minimum: item ID, title, description, URL, brand, seller name, image URL, availability and price. It can be sent as JSONL, CSV or TSV, and price and availability should be updated when they change. Buying inside the chat is a separate integration that has to be enabled; since March 2026, OpenAI focuses on discovery and merchants use their own checkout.

What do I need for Google agentic checkout?

An active Merchant Center account, approved products, declared return policies, and a marker in the feed for each product eligible for checkout. As of October 2026 it is available to US merchants, and the Universal Cart has been live in US Search and the Gemini app since October 15, 2026.

Do I need an llms.txt file?

It is not a requirement for appearing in shopping assistants, and it doesn't replace structured data or feeds. If you have everything else solved, adding it does no harm; if you don't, it is not the place to start.

Is Shopify already ready for agentic commerce?

Shopify covers a good part of the technical side: structured data, feeds and APIs, plus a sales channel that publishes your catalog to assistants such as ChatGPT, Copilot and Perplexity, with availability that varies by assistant. WooCommerce and BigCommerce cover the basics too, with more of the work in plugins, apps and configuration. On every platform, what depends on you is data quality, synchronized inventory and clear policies.

Should I block AI bots from my store?

It depends on what you want. If you block the assistants' crawlers, your products have fewer chances of appearing in their answers. It is worth distinguishing between known assistant crawlers, which you can allow, and abusive automated traffic, which your fraud tool should keep stopping.

What happens if an agent buys the wrong thing from my store?

Your return policy applies, just as with a person. That is why the attributes that prevent mistakes (size, compatibility, dimensions) should be clear on the product page, and the return policy should be published and structured.

Where do I start if I'm short on time?

With reliable inventory and price, and a Merchant Center account with no errors. Those are the two points that weigh the most today for appearing in search engines and assistants, and the ones that cause the most trouble when they fail.

If you want us to review how AI agents read your store and what it would take to connect an agent of your own for support and reorders, we build a free working demo in 24 to 48 hours. You can book a call.