AgentLens · Guide
How AI agents see your Shopify store
What an AI shopping agent or answer engine actually fetches from your storefront, what it cannot see, and three ways to check your own store.
When a shopper asks an AI assistant “does this shop ship to Ireland?”, the assistant does not browse your store the way a person does. It fetches a handful of URLs, reads the text and the machine-readable data on them, and answers from that. If the answer is not in those fetches, it guesses, hedges, or recommends someone else.
What an agent actually fetches
- robots.txt. Every Shopify store serves one by default, and a theme can replace it with its own
robots.txt.liquidtemplate. It holds groups of rules per user agent saying which paths may be fetched, plus an optional sitemap URL. AI crawlers announce their own user-agent strings, so a rule aimed at them (or a blanket disallow) decides first whether they see anything at all. - The sitemap. Shopify generates
/sitemap.xmlautomatically and updates it when you add a product, page, collection, image or blog post. It links to every product, its primary image, pages, collections and blog posts. For an agent this is the catalog index. - Product pages and their structured data. The HTML title, meta description, headings and description text, and, if the theme emits it, a
Productblock of JSON-LD with the name, offers, price, availability and image. Shopify's documentation does not promise that every theme includes this block, so check yours rather than assume. When it is there, it is the highest-signal input an agent gets about a product. - Product JSON. Shopify's Ajax API documents
GET /products/{handle}.js, which returns a product's JSON: title, variants, prices, images and options. Many storefronts also answer/products.jsonwith a listing, but that form is not in the reference, so try it rather than rely on it. An agent that finds either gets a price without parsing your theme's HTML. - Agent discovery files and endpoints.
llms.txtandagents.mdare conventions, not Shopify features: a plain-text summary of the site and a note on how an agent should interact with it. Shopify is also building agent-facing endpoints for catalog search and carts, and this area is moving: its documentation now points developers from the earlier storefront MCP endpoint to a newer Universal Commerce Protocol endpoint. What your store exposes today depends on what Shopify serves and what you have enabled, so test it rather than assume it.
Sources: Shopify Help Center, “Editing robots.txt” and “Finding and submitting your sitemap”; shopify.dev, robots.txt.liquid template reference, Ajax Product API reference, and “Migrate from Storefront MCP”.
What it cannot see
- Anything behind the password page. In private mode all of your pages are hidden from visitors and search engines and the password page is shown instead; Shopify notes that a sitemap reader cannot access the sitemap in that state. A password-protected store has, to an agent, no products. The same goes for anything shown only to a logged-in customer.
- Images without alt text. A text fetcher gets an image's
altattribute and nothing else. A product with six photos and no alt text is, to the agent, a product with no photos. - Claims that exist only in an image. “Free shipping over $50” baked into a banner, a size chart that is a JPEG, a certification badge: none of these are in the text the agent reads. If the claim matters, put it in the page text or the policy page as well.
- Content that only appears after scripts run. Tabs and accordions filled in by JavaScript may be skipped by a fetcher that does not execute it.
Method 1 — by hand, with curl or view-source
Free, ten minutes, and it tells you most of what you need. In a terminal, with your own domain and handle:
curl -s https://your-store.com/robots.txt curl -s https://your-store.com/sitemap.xml | head -40 curl -s https://your-store.com/products/your-product-handle.js curl -s https://your-store.com/llms.txt
Read robots.txt for any Disallow rule that covers /products or names an AI crawler, and check that a Sitemap: line is present. Confirm the sitemap is an index of product URLs, not a password page. The .js endpoint should return JSON with the title, variants and prices; HTML means the store is private or the path is wrong.
Then open a product page, use View page source, and search for application/ld+json. Look for a block whose @type is Product, with name, offers (price and currency), availability and image. A missing block or an offer with no price is your first fix.
Method 2 — a rich-results test and a text-only view
The major search engines each publish a free rich-results or markup testing tool. Paste a product URL and it reports which structured data it detected and every error or missing field. A clean result here is worth having regardless of what you do next.
Then read the page the way a text-only fetcher does: reader mode, a browser with images and JavaScript off, or curl piped through any HTML-to-text tool. Read what is left as a shopper with a question. Is the material named? Is the shipping policy a sentence or a picture? What disappeared is what the agent never had.
The gap in both methods is that they check one page at a time, and structure rather than answers: valid markup, not whether “do you have this in a size 12?” gets a correct reply from your catalog.
Method 3 — an app that runs the audit and drafts the fixes
This is what AgentLens does. It generates shopper questions from your own catalog and works through your storefront the way an agent would: discovery files, robots.txt rules for AI crawlers, the sitemap, product JSON-LD, the Organization schema on your homepage, image alt coverage, the agent endpoint, catalog search, policy questions about shipping and returns, and whether a cart can be created and handed to checkout. Every finding shows the exact transcript, why it passed or failed, and the fix. You get a 0 to 100 agent-readiness score and a separate search-engine SEO score (titles, meta descriptions, indexability, canonicals, headings, Open Graph tags, robots.txt, mobile viewport, response time), kept apart because the two questions are different.
The Products & pages screen gives every product and page two verdicts, one for AI agents and one for search engines, with the concrete issues on each. Store-level findings come with a guided fix naming the admin screen or theme file. For content, the app drafts the rewrite beside the original; nothing publishes until you click Approve & apply, and any change can be reverted. Audits are read-only, and the app requests no customer, order or checkout data.
Pricing, billed through Shopify: Free is $0/month for 3 products and pages, with a weekly audit and guided fixes. Growth is $39/month for 50, with scheduled audits, email digests and AI-drafted rewrites. Pro is $99/month for 1,000, adding the bulk optimizer. Enterprise is $499/month, unlimited, with 4,000 optimizations a month included and $0.20 each after. There is no free trial; the Free plan is the trial.
Which one to pick
If you have a short catalog and a few minutes, method 1 is a real answer and the commands above are complete: a blocked crawler, a private store, a missing sitemap line or a product with no price in its markup are the fixes that matter most. Method 2 is the right step whenever you touch the theme; it catches malformed structured data a visual check never will, and the text-only read is the most useful habit on this page. The app earns its price when you have more products than you will check by hand, when you want the shopper-question probes and the cart test rather than just the structure, or when you want the rewrites drafted rather than described. If you are not sure, the Free plan audits three targets for nothing, and the first two methods cost nothing at any size.
Questions
Which files does an AI agent read first on a Shopify store?
robots.txt and sitemap.xml. Every Shopify store serves both by default: robots.txt says which paths a crawler may fetch and can point at the sitemap, and the sitemap lists products, the primary product image, pages, collections and blog posts. From there an agent fetches product pages and reads their text and structured data.
Can an AI agent see a password-protected Shopify store?
No. In private mode every page is hidden from visitors and search engines and the password page is shown instead, and Shopify's own documentation says a sitemap reader cannot access the sitemap while the store is private. An audit of a password-protected store reads as unreachable for the same reason.
Does an AI agent read the text in my product images?
Not reliably. A fetcher that reads text gets the alt attribute of an image and nothing else, so an image without alt text is invisible and a claim that only exists inside a banner graphic does not exist for the agent. Put the claim in the page text as well.
How can I check how my Shopify store looks to an AI agent for free?
Fetch robots.txt, sitemap.xml, a product page and that product's .js endpoint with curl or view-source and read what comes back; then paste a product URL into a search engine's rich-results test and read the page in a text-only view. That covers the structure. An audit app adds the shopper-question probes and scoring, and drafts the fixes.
Does AgentLens change my store when it audits it?
No. Audits are read-only. Content changes happen only when you approve a draft, and every applied change can be reverted. The Free plan is $0 per month and covers three products and pages; there is no free trial because the Free plan is how you try it.