By Steve Merrill, Founder of WRKNG Digital | September 20, 2026
A shopper doesn't type a search anymore. She tells ChatGPT to find a rain jacket under $150 with free returns and buy it. The assistant does the whole thing. Browses, compares, decides, checks out.
That's agentic commerce. And most Shopify stores break the agent somewhere in that flow without ever knowing it happened.
I've had a hard time getting store owners to care about this early, because the buyer never sees the break. There's no bounce report for an agent that gave up. It just moves on to the store it could read. So let me walk you through exactly what an AI shopping agent does, step by step, and show you where yours is losing.
What is an AI shopping agent, and how is it different from search?
An AI shopping agent is software that shops and buys on a person's behalf. It reads product data, compares options across stores, and completes checkout inside an assistant like ChatGPT or Google's AI experiences. Search hands a person links. An agent takes the action.
That difference changes everything about how you prepare your store. Search rewards pages people click and read. An agent rewards data it can read and trust. OpenAI announced Instant Checkout in ChatGPT in late 2025, built on the open Agentic Commerce Protocol it developed with Stripe. Google has been pushing its own agentic shopping through Shopping Graph and AI Mode. Shopify plugged into these rails so merchants wouldn't have to build them alone.
Here's the bottom line. The buyer talks to the assistant. The assistant talks to your store's data. If your data is a mess, you're not in the conversation.
How does an AI shopping agent browse and buy, step by step?
An AI shopping agent moves through five steps: discovery, comparison, evaluation, add-to-cart, and checkout. At each step it reads a specific piece of your store's data and decides whether to keep going or drop you. A store that wins keeps the agent moving through all five without friction.
Let me break down each step and the exact place stores break.
Step 1: Discovery - can the agent even find your product?
In discovery, the agent reads structured product data and feeds to build a list of candidates that match the request. It's not reading your pretty product page. It's reading the data behind it: title, price, availability, category, key specs. No clean data, no candidacy.
This is break point number one. Missing or thin structured data.
Plenty of Shopify stores ship product pages with no Product schema, or schema that's missing price and availability. The agent hits the page, finds nothing it can parse, and moves on. You never made the list.
Fix it: Add valid Product structured data to every product page with name, price, priceCurrency, availability, and brand at minimum. Feed the same clean data into your Shopify product feed so Google's Shopping Graph and other crawlers get it too. Test it in Google's Rich Results Test before you trust it.
Step 2: Comparison - can the agent line you up against rivals?
In comparison, the agent pulls price, shipping cost, delivery speed, return terms, and core specs from every candidate and stacks them side by side. It's building a spreadsheet in its head. Whatever field it can't read for your store shows up blank next to a competitor who filled it in.
Break point two: gated or buried information.
Shipping cost hidden until checkout. Return policy locked inside a PDF. Material and dimensions written into an image instead of text. A human will dig for that. An agent won't. It reads what's readable and scores you on what's there. Blank fields lose.
I saw this on a client's store last quarter. Great product, better price than the competitor, and the agent kept recommending the rival because our shipping and return terms lived in a support-center article the crawler never opened. We moved that data onto the product page as text and structured markup. The recommendations flipped.
Fix it: Put shipping cost, delivery window, and return terms in plain text on or near the product page. Add shippingDetails and merchantReturnPolicy structured data. Get specs out of images and into readable copy. Assume the agent only sees text and data.
Step 3: Evaluation - can the agent trust you enough to pick you?
In evaluation, the agent checks reviews, ratings, stock status, and policy details to rank the shortlist and choose. This is where trust signals decide the winner. An agent won't buy something it can't confirm is in stock, and it leans hard on review data to break ties between similar products.
Break point three: stale or unreadable availability and reviews.
If your availability field says in stock but the variant is actually sold out, the agent may pick you and then fail at checkout. That's worse than being skipped. And if your reviews live in a third-party widget that renders as JavaScript with no structured data, the agent scores you as having none. You look weaker than you are.
Fix it: Keep availability accurate at the variant level and expose it in your schema and feed. Surface aggregateRating and review structured data so the agent can read your social proof. Sync inventory in real time so the agent never picks a product you can't sell.
Step 4: Add to cart - can the agent choose the right variant?
In add-to-cart, the agent selects the exact variant the buyer wants and adds it. Size, color, material, quantity. It needs a clean, machine-readable variant structure to map "medium blue" to the right SKU. Ambiguous variants are where confident agents get confused and quit.
Break point four: ambiguous variants.
This one's everywhere. Options labeled "Style 1" and "Style 2" with no meaning. Colors named "Ocean" and "Storm" with no plain color attached. Two variants with the same title. The agent can't tell them apart, so it either guesses wrong or refuses to add anything. Either way the sale dies in the cart.
Fix it: Name variants in plain, literal terms. Map every variant to a real color, size, and material attribute in your product data, not just a marketing name. Give each SKU a unique, unambiguous title. If a human needs the size chart to decode your options, an agent is already lost.
Step 5: Checkout - can the agent actually complete the purchase?
In checkout, the agent completes payment through a supported flow. With OpenAI's Instant Checkout and the Agentic Commerce Protocol, the buyer confirms and pays without leaving the assistant. If your store doesn't support an agent checkout path, the agent hits a wall at the last step and the whole journey was wasted.
Break point five: checkout friction the agent can't clear.
Forced account creation. A pop-up asking for an email before checkout. A multi-step flow that assumes a human is clicking. Agents handle clean, supported checkout rails well. They stall on the surprise friction you added to capture leads from human shoppers.
Fix it: Turn on the agentic checkout support Shopify offers rather than building your own. Shopify has been rolling out merchant support for these buy-in-assistant flows, so check your admin and your app options. Strip mandatory account creation and interstitial pop-ups from the buying path. Make the last step something an agent can finish.
Why should a Shopify owner care about this in 2026?
Because the agent picks the store, and the buyer trusts the pick. When a shopper tells an assistant to buy the best-rated option under budget, she rarely audits the five stores it compared. She takes the recommendation. That puts the store with the cleanest, most complete data in front of the sale.
Two percent of your traffic being agents today feels ignorable. It won't stay that small. OpenAI, Google, and Shopify are all building the rails at the same time, which tells you where the buying is headed. The stores that fix their data now get recommended while their competitors are still invisible to the agent.
I'm not promising you a revenue number. Nobody honest can. What I can tell you is which break points cost you the recommendation, and every one of them is fixable this month.
How do I make my Shopify store agent-ready?
Work the five break points in order. Add clean Product schema and a full feed for discovery. Expose shipping, returns, and specs as text and structured data for comparison. Keep availability and reviews accurate and readable for evaluation. Fix ambiguous variants for add-to-cart. Turn on supported agentic checkout and cut friction for the final step.
Start with structured data and variants. Those two break the most agents and they're the fastest to fix. Then work backward from checkout, because a broken last step wastes every step before it.
If you want the full audit against the Agentic Commerce Protocol and Google's agentic shopping requirements, that's the work we do at WRKNG Digital. See how we get Shopify stores agent-ready here.
FAQ
Can an AI agent actually check out on a Shopify store?
Yes, when the store supports an agent-ready checkout. OpenAI's Instant Checkout and the Agentic Commerce Protocol let an agent complete a purchase inside the assistant. Shopify has built support for these flows, so a merchant on Shopify can turn them on rather than build from scratch.
What breaks an AI shopping agent most often?
Missing structured data and ambiguous variants break agents most often. If the agent can't read your price, availability, and specs as clean data, or can't tell your medium blue shirt from your large black one, it drops your product before it ever reaches checkout.
Does good SEO mean I'm ready for AI agents?
Not on its own. SEO ranks pages for people who read and click. An agent reads data and acts. You need Product schema, a clean product feed, plain-text shipping and return terms, and a supported checkout path before an agent can buy from you.
Where do I put shipping and return information for agents?
Put it in structured, machine-readable form near the product, not buried in a PDF or a chat widget. Use Schema.org shippingDetails and merchantReturnPolicy, and keep a plain-language version on the page. Agents skip data they can't read cleanly.
