Is your Shopify store readable by AI shopping agents? A checklist
ARLing s. r. o. · Sources checked 13 October 2026
A shopping assistant needs enough reliable information to distinguish your product from an alternative: what it is, which variant is available, what it costs and what happens after purchase. Start by checking the information a visitor can actually read. No checklist can guarantee a recommendation or a sale.
Shopify documents product discovery through Shopify Catalog. A public /products.json response is only one outside view of a store; it is not proof of the data or channel settings in your private catalog. Shopify Catalog and product discovery.
1. Read a real product, not just your home page
Choose a product with variants and open its page while signed out. Record the URL and the time. Check whether the title, material, dimensions, intended use and variant differences are understandable without a marketing slogan.
For Shopify, compare your public export with the product record in your admin. Product type, vendor and option names are useful fields to inspect. An empty public field is evidence about the export, not necessarily about the underlying catalog. Shopify product object.
2. Make variants distinguishable
Two variants called “Default” do not help someone choose a size. Check that each option has a name and that each variant has the appropriate values, price and availability. Sold out is a valid state. A zero price can be valid too; it should be intentional.
If the manufacturer assigned a GTIN, use that identifier accurately. A valid checksum does not prove that the identifier belongs to the product. If a public feed omits the barcode field, inspect the admin record before trying to “fix” it. Never invent a GTIN to satisfy a checker. Shopify variant fields.
3. Inspect the data behind the visible offer
Look for Product and Offer JSON-LD in the product page HTML. The structured name, image, price, currency and availability should agree with what a shopper sees. If your theme describes a group of variants, inspect those relationships too.
Delivery and return information can be represented in structured data, but some merchants configure it in their platform account instead. An outside checker cannot inspect those private settings. Use Google's validation tools and documentation when targeting Google's merchant listings. Do not treat a home-made score as a platform approval. Google merchant listing documentation.
4. Make the buying conditions findable
Follow your own return, shipping and contact links. Do they open without login? Do they contain the policy for this shop, rather than an empty template? Can a visitor work out where you deliver and how to contact you?
An HTTP 200 response proves only that a server answered. It does not prove a policy is complete, lawful or accurate. Read the page as a customer and keep it consistent with your checkout.
5. Separate search access from model training
There is no single “allow AI” switch. Start with the exact user-agent token and its documented purpose.
| Token | What to check |
|---|---|
| OAI-SearchBot | ChatGPT search crawling preferences. |
| ChatGPT-User | User-triggered requests; robots.txt may not govern them. |
| PerplexityBot | Perplexity search crawling, separate from model training. |
| Google-Extended | Gemini training and certain grounding uses, not Google Search inclusion or ranking. |
Check the actual product path, not only /. A longer Allow rule can change the result of a broader Disallow. Passing this inspection still does not show whether a firewall permits a platform's real network requests. Sources: OpenAI crawlers, Perplexity crawlers, Google crawler controls, robots exclusion protocol.
6. Treat optional signals honestly
llms.txt is a proposal for a concise guide to site content. It is not a shopping eligibility certificate. Publishing one is optional; do not postpone product or pricing corrections to chase it. The llms.txt proposal.
Ratings are also not something to manufacture for a better score. If you publish AggregateRating, make sure real reviews support the value and count and are accessible to shoppers. A parser can check numeric consistency; it cannot authenticate the customers. Google review guidance.
7. Keep a reproducible record
Save the sampled URLs, timestamp, result and limits. After a theme change, compare the same products again. A different sample can change a score without any underlying improvement. If a page times out, record it as unverified and retry; do not mark it correct.
A helper for the public checks
Feed Doctor's Agent Ready mode reads a limited public sample and lists gaps, sources and suggested corrections. It shows coverage beside its heuristic score. It cannot inspect your private catalog, guarantee visibility or verify checkout. Start with the free check; use its findings as a review list.