How the AI-Readiness Checker Works (and Why We Built It)

We built a free AI-readiness checker because most Shopify merchants have no way to answer a question that is starting to matter a lot: when a shopper asks ChatGPT, Claude, or Perplexity for a recommendation, can those assistants actually read your store?

Google has the Rich Results Test for its own features. There was nothing equivalent for the new wave of AI shopping agents. So we made one, and this post explains exactly how it works, because a tool that grades your store should be willing to show its own work.

What it reads

When you paste a URL, the checker fetches your store the way an AI agent would: a plain bot user agent, a short timeout, and no JavaScript execution. It pulls five things:

1. Your homepage HTML.
2. One representative product page. The tool scans your homepage for a real /products/ link and follows it, because the signals that decide whether an agent can recommend you live on product pages, not the homepage.
3. Your /llms.txt, the index file AI assistants increasingly look for.
4. Your /robots.txt, to see whether AI crawlers are allowed.
5. Your /sitemap.xml, so crawlers can discover every page.

It reads, scores, and closes the connection. Results are not stored, and there is no sign-up.

The five dimensions

The checks are grouped into five dimensions, each weighted by how much it affects whether an agent can find, understand, and recommend you.

1. Brand identity

Can an agent tell who you are? This looks for Organization or WebSite structured data, a logo published in that schema, linked social profiles (sameAs), and a SearchAction that lets engines search into your catalog. Without these, an assistant cannot reliably connect your store to your brand.

2. Products and catalog

This is the heart of it. The tool checks the product page for Product (or ProductGroup) schema, an Offer with a price and availability, AggregateRating from reviews, and BreadcrumbList for catalog structure. Product and Offer data carry the most weight in the score, because an agent that cannot read your price and availability cannot put you in an answer.

3. AI discovery files

A valid /llms.txt, an AI-friendly /robots.txt, and an XML sitemap. For robots, the checker reports each major AI crawler by name: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, and Applebot-Extended. If your robots file blocks them, agents never see your content in the first place.

4. Answer-engine basics

The plain web fundamentals answer engines summarize from: a descriptive page title, a meta description of useful length, a canonical URL, complete Open Graph tags, and a declared page language. These are not glamorous, but they are what an engine quotes when it cites you.

5. Trust and policies

The signals that make an agent confident enough to recommend you: a machine-readable return policy (hasMerchantReturnPolicy), structured shipping details, and FAQPage schema. When a shopper asks "can I return this?", an agent can only answer if the answer is in the data.

How scoring works

Each check resolves to one of three states: pass, warn, or fail. A pass earns full weight, a warning earns half, and a failure earns none. The weighted total becomes a score from 0 to 100 and a letter grade from A to F. Each dimension also gets its own score, which is why the result shows a radar across all five rather than a single number. A store can be excellent at products and weak at identity, and the breakdown shows you where to spend your effort.

"What AI sees about you"

The most useful panel in the result is not a score. It is the plain-language summary of what an assistant could actually extract: your store name, what you sell, whether your products are machine-readable, your price, your rating, and your linked social profiles. When that panel is mostly blank, it is showing you something important. Your store might look great to a human and be almost invisible to an agent reading structured data.

We ran the checker against a few well-known stores while building it. One large, beloved brand scored well on products but had no Organization schema at all, so the "what AI sees" panel could not even confirm the company name from structured data. That gap is invisible in a browser and obvious to a parser.

Every finding comes with a fix

The point of the tool is not the grade. It is the list of fixes. Every warning and failure includes a concrete, do-it-yourself fix written in plain language. Where AgentReady can resolve a gap automatically, we mark it clearly, so you always know the difference between "here is something to do" and "here is something we can do for you."

If you want the score to go up, run your store through the checker, fix the failures first and the warnings next, and re-run it. If you would rather not hand-edit schema, that is exactly what AgentReady is for: it publishes the structured data, the llms.txt index, and the policy signals for you, and keeps them in sync as your catalog changes.