How to Check If AI Shopping Assistants Can See Your Shopify Store

A merchant wrote to us last month with a tidy summary of the problem: "AI shopping assistants are not finding my store, and I have no idea how to even check why." Fair. There is no dashboard in your Shopify admin called Agent Visibility, and nothing errors when an assistant skips you. Your store looks fine to you and to Google, and meanwhile ChatGPT recommends three competitors.

The good news is that "can AI see my store" is not a mystical question. It decomposes into five concrete checks, each one maps to a layer that can fail, and you can run all five in about ten minutes with a browser. Here is the walkthrough.

Check 1: Are AI crawlers allowed in?

Open yourstore.com/robots.txt in a browser. This file is the bouncer for every automated visitor, and it now governs the AI crawlers too: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and friends.

Scan for two things. First, any block that names one of those agents with a Disallow: / under it. Second, a blanket User-agent: * disallow that an old SEO app or a copy-pasted snippet left behind. Either one means the assistants' crawlers never reach your product pages, and nothing downstream can fix that.

Shopify's default robots.txt is workable, but stores that customized it years ago sometimes inherited rules written for a different era. The full list of which agents to allow, and what Shopify's default actually does, is in which AI crawlers to allow in your robots.txt.

Pass: no AI crawler is disallowed. Fail: fix this first, because every other check depends on it.

Check 2: Does your store introduce itself?

Open yourstore.com/agents.md. Every Shopify store now serves this file at the domain root. It is the brief an AI agent reads to learn what your store is, what it sells, and how to transact with it, and on Shopify it is also mirrored at /llms.txt.

The check here is not "does the file exist," because on Shopify it always does. The check is whether it says anything. Read it as a stranger. Does it describe your actual brand, your categories, your policies? Or is it the generic platform default that reads identically to a million other stores? A default file is technically present and practically silent, and we have covered why that costs you recommendations in your agents.md is identical to every other store.

Pass: the file names your brand and describes what you sell. Fail: it exists but says nothing specific, which means agents have to guess.

Check 3: Can an agent extract hard facts from a product page?

Assistants do not recommend vibes. They recommend a product at a price, in stock, with ratings, and those facts come from the structured data on your pages. Open one of your best-selling product pages, view the page source, and search for application/ld+json. You are looking for a Product block that carries price, availability, and ideally review ratings. Google's Rich Results Test does the same inspection with less squinting.

Most Shopify themes emit some Product schema, but coverage is uneven: variants missing, availability absent, ratings locked inside an app widget the schema never mentions. The full anatomy of what a complete block looks like is in the product schema guide.

Pass: a Product block with price and availability on your key pages. Fail: an agent that reaches your page still cannot quote it confidently.

Check 4: Are your products in the catalog assistants actually query?

Here is the layer most merchants have never heard of. When ChatGPT, Gemini, Copilot, or the Shop app answers a shopping question against Shopify's ecosystem, it is not crawling your storefront in the moment. It queries Shopify's Global Catalog, a cross-merchant index built from every store's product data. If your products are absent from that index, or filed under a category Shopify's inference guessed wrong, you are invisible in exactly the surfaces that matter, while your storefront looks perfect.

The quick probe: open the Shop app and search for your product by category, not by brand. If a search that should obviously include you returns competitors and nothing of yours, the catalog layer deserves a hard look. The full diagnosis of the three ways this fails, absence, mis-categorization, and stale data, is in why your Shopify products don't show up in AI shopping.

Pass: your products appear for category searches in Shop. Fail: the problem lives in the catalog, not on your website.

Check 5: Ask the assistants like a stranger

Finally, the customer's-eye view. Open ChatGPT and ask the question a new customer would ask. Never your brand name, always the category: "a good ceramic pour-over dripper under $50," whatever your version is. Note whether you appear, where, and who beat you. Repeat in Gemini and the Shop app if you have another few minutes.

One answer is a data point, not a verdict. Assistants vary their answers, and a single miss means little. But if ten category questions across two assistants never once surface you, that is signal, and the checks above tell you which layer to suspect. If you want to make this a repeatable monthly habit with a scorecard, how to tell if you're winning or losing in the Shopify Catalog turns it into a system.

Reading your results

The five checks stack in order. Blocked crawlers make everything else moot. A silent agents.md leaves agents guessing what you are. Missing structured data leaves them unable to quote you. Catalog gaps hide you from the biggest assistant surfaces regardless of your website. And the direct questions tell you what shoppers actually experience.

None of this guarantees a ranking, and be suspicious of anyone who says otherwise. What the checks establish is eligibility: whether an assistant that wants to recommend you can find you, understand you, and trust your details. Most stores that feel invisible are failing one of these five, and it is almost never the one they guessed.

If you would rather not do the manual round, the free AI-Readiness Checker runs the same inspection the way an agent would, grades each layer, and shows you exactly what to fix first. It takes about a minute, which is nine minutes better than this post.