Your AI Assistant and Your Shoppers' AI Read the Same Catalog

There are two groups of AI assistants reading your Shopify store right now.

The first group works for you. Sidekick in your admin, or a Claude or ChatGPT you've connected, answering your questions about stock and sales.

The second group works for your shoppers. Someone asks ChatGPT, Gemini, or Perplexity for "a carry-on that fits under a seat," and the assistant goes looking through stores, yours included, to decide what to recommend.

You set up the first group, so you think about it. You never see the second group, so it's easy to forget. But they have something important in common: both are reading the same product data. Get that data right and both get better. Leave gaps and both get worse, though only one of them will tell you.

Same data, different doors

Your assistant comes in through the front office. It signs in with your permissions and reads the admin directly: every field, every variant, every inventory count at every location.

A shopper's assistant comes to the shop window. It reads what your store publishes: your product pages, the structured data in them, files like agents.md and llms.txt, and what Shopify's catalog holds about you. We go through that view in detail in what AI assistants see when they read your store.

That last row is the whole reason for this post.

A thin answer is a warning

Say you ask your assistant: "What's the Atlas backpack made of, and will it fit a 16-inch laptop?"

If your data is good, it answers in a second: recycled nylon, 22 litres, laptop sleeve fits up to 16 inches.

If your data is thin, you get something like: "The description mentions it's durable and great for commuting. I don't see specific material or laptop size details."

The second answer is annoying. It's also a free diagnostic. Your assistant has more access to your store than any outside AI will ever have. If it can't find the laptop size, a shopper's assistant certainly can't. And a shopper's assistant doesn't say "I don't see those details." It recommends the backpack from the store that stated them.

So every vague answer from your own assistant is pointing at a gap that's costing you somewhere you can't see.

A ten-minute test

Pick one of your best sellers. Ask your assistant the questions a shopper would ask before buying. Not sales questions. Product questions.

• "What is [product] made of?"
• "What sizes and colours does it come in, and which are in stock?"
• "What are its dimensions and weight?"
• "Who is it for, and what is it not suitable for?"
• "How long does shipping take, and what's the return policy?"
• "How is it different from [your similar product]?"

Score each answer:

• Direct and specific: the fact lives in a structured field. Good.
• Correct, but dug out of a paragraph: the fact exists only in prose. A person can find it. A machine may not.
• Vague or missing: the fact isn't in your store at all.

Do this for three products and the pattern in your catalog will be obvious. It's usually the same two or three fields missing everywhere.

Passing your own test isn't the finish line

Your assistant sees the admin. Shoppers' assistants see the storefront. A fact can be perfect in one and absent from the other.

Common ways that happens:

• The fact is in a metafield your theme doesn't output. It's in your admin and nowhere on the page.
• Your structured data is incomplete. The page shows the price and stock to a person, but the machine-readable version is missing them.
• The published data is stale. The admin says out of stock. The data an outside assistant last read says available. We cover this in why AI shows the wrong price or availability.
• Reviews live in a widget. Your star rating shows to people and isn't in the page's data. See can AI assistants see your product reviews?
• A crawler is blocked. The data's fine and the assistant isn't allowed in. See AI crawlers and your Shopify robots.txt.

So there are two checks. The inside check, with your own assistant, tells you whether the facts exist. The outside check tells you whether they're published.

What to fix first

In rough order of payoff:

1. Titles that say what the thing is. "Atlas" is a name. "Atlas 22L Commuter Backpack" is a product.
2. Product type and category. These are how a product gets matched to "backpack" in the first place. See product taxonomy as a ranking factor.
3. Variants with real option names. Size and Colour, with values a shopper would say.
4. Specs in structured fields, not only in the description.
5. Identifiers. GTINs where your products have them. See GTINs and product identifiers.
6. Descriptions made of facts. Materials, dimensions, and uses. Fewer adjectives.
7. Policies as readable content. Shipping times and return windows, stated plainly. See returns and shipping policy data.

AI can do much of this work with you. The safe way, including which fields it must never guess, is in filling missing product data with AI. The fuller list of what carries weight is in the product fields that affect AI shopping rank.

One piece of work, paid twice

This is the encouraging part. You don't have two projects, one for your own tools and one for AI shopping. You have one.

Fill in the laptop size on the Atlas backpack and:

• Your assistant answers the support question correctly.
• A storefront filter can use it.
• A shopper's assistant can match it to "fits a 16-inch laptop."
• Google has one more fact for its own listings.

Clean product data has always been worth having. What's changed is how many readers it now has, and how many of them are deciding whether to send you a customer.

Run both checks

The inside check is the ten-minute test above. Use Sidekick, or connect an assistant with our guides for Claude and ChatGPT.

The outside check is the free AI readiness checker. It fetches your store the way an AI assistant would and reports what it could and couldn't read. It takes about a minute.

Closing the gap between the two is what AgentReady does: it publishes and maintains the structured data and discovery files that make what's in your admin readable from the outside.

For everything your own assistant can do once it's connected, see every AI option for running a Shopify store.