Is your Shopify store invisible to AI? The five failure points

Shopify merchants tend to assume the platform has this covered, and since May 2026 it partly does. That is exactly what makes a Shopify store invisible to AI harder to spot than a WooCommerce one: the obvious layer looks handled, so the remaining failures sit lower down, in the theme and in the app stack, where nobody is looking. Agent readiness is scored across five weighted categories, and on Shopify each one fails in its own specific way.

The reason this became urgent is covered in the switch was flipped. The short version: AI-referred traffic and AI-attributed orders grew sharply through 2025 and 2026, and agents pick merchants partly on what they can read without a human involved.

1. Discovery (20 percent): usually a partial pass

Discovery covers llms.txt, llms-full.txt, agents.md, a UCP manifest at /.well-known/ucp, a valid sitemap, and whether that sitemap actually exposes product URLs.

Shopify ships llms.txt, agents.md and Universal Commerce Protocol support natively as of May 2026, so on many stores several of these checks now pass with no work from you. Two caveats. Native support is not the same as universal coverage: Shopify enabled agentic storefronts for eligible merchants in March 2026, and eligibility is Shopify's determination, not something to infer from your plan. And a generated file only helps if it describes the catalogue you want recommended.

So look instead of assuming:

curl -s https://yourstore.com/llms.txt | head -40
curl -s -o /dev/null -w "%{http_code}\n" https://yourstore.com/.well-known/ucp

A 404 means you are not covered. If content comes back, read it and judge whether it points at your real collections and products. llms.txt for Shopify goes through the file properly. These are the discovery.llms_txt, discovery.agents_md and discovery.ucp_manifest checks.

2. Crawl access (25 percent): the robots.txt.liquid trap

Crawl access is the heaviest category alongside product schema, and it behaves close to binary. If a crawler is blocked, nothing else in the report matters for that agent. Five are checked by name: GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot and Google-Extended.

Do not reason about this from the platform default. Read the file your store actually serves, because that is the only version that counts. Shopify allows a robots.txt.liquid template, and customisations there are where accidental AI crawler blocks come from, often added years ago to deter scrapers, or written by an SEO app during setup. The template is easy to forget because it is not in Settings, it is in the theme code editor.

Check the rendered file rather than the template:

curl -s https://yourstore.com/robots.txt | grep -iE "gptbot|claudebot|oai-searchbot|perplexitybot|google-extended|disallow: /$"

Anything that comes back deserves a deliberate decision. Robots.txt for ecommerce and AI crawlers covers how to allow retrieval bots while still blocking what you want blocked. The same category checks HTTPS, reachability and a homepage response under three seconds, and on Shopify the last of those is the one that slips, usually from app scripts and hero video rather than from hosting.

3. Product schema (25 percent): whatever your theme emits

This is theme-level, not platform-level. Shopify does not guarantee complete Product JSON-LD on your product pages; your theme's product template does, and apps can overwrite it.

The base is usually present: a Product object with an Offer carrying price, currency and availability. What goes missing more often is sku, gtin or mpn, aggregateRating and BreadcrumbList. AggregateRating deserves separate attention because it does not come from your theme at all. It comes from your reviews app, and plenty of review widgets paint stars into the page after load without ever emitting structured data, which fails product.aggregate_rating even though a shopper can see the rating clearly. A minimally complete shape looks like this:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Cedar Bath Tray",
  "sku": "CBT-001",
  "gtin13": "0123456789012",
  "offers": {
    "@type": "Offer",
    "price": "48.00",
    "priceCurrency": "GBP",
    "availability": "https://schema.org/InStock"
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.7",
    "reviewCount": "212"
  }
}

The schema fields AI agents actually read explains why identifiers like gtin matter for matching your product to a query.

4. Merchant data (15 percent): policies written as prose

MerchantReturnPolicy, OfferShippingDetails, FAQPage and Organization sit here. Agents select merchants partly on structured data completeness, and these properties are specifically part of that judgement.

The Shopify version of this failure is almost universal in shape. Returns and shipping live in Settings as policy pages, and those pages render as rich text. A human reads them fine. An agent comparing you against another merchant cannot dependably pull a returns window, a restocking condition or a delivery estimate out of a paragraph, so it uses the merchant that stated those terms as fields. Add MerchantReturnPolicy with returnPolicyCategory, merchantReturnDays and returnMethod, and OfferShippingDetails with a shipping rate and a delivery time, then keep the values identical to the prose page so the two never disagree.

5. Content quality (15 percent): supplier copy and hidden images

The checks are plain: product titles of at least 30 characters, descriptions of at least 500 characters, and three or more detectable images.

Two Shopify patterns fail this. Catalogues imported from a supplier feed inherit short generic titles and one-line descriptions, which give an agent almost nothing to match a query against. And image galleries built by an app sometimes hold every shot after the first in JavaScript, so only one image is detectable in the HTML that a crawler receives. View source on a product page and count the image URLs you can actually see.

If you want to know which of these five your own store fails rather than which ones it probably fails, run a free scan. It reads only public pages and takes under a minute.