How to make your Shopify store readable for ChatGPT, Claude and Perplexity

An AI-readable store is one where every fact an assistant needs — price, availability, variants, materials, shipping, returns — is published as machine-readable data instead of being locked inside images, scripts or unstated assumptions. When someone asks ChatGPT, Claude or Perplexity "where can I buy a linen duvet cover that ships to Austria?", the assistants that answer well do it by reading stores — their structured data, their policies, their product pages. This guide is a seven-step checklist for making your Shopify store one of the stores they can actually read. One honest note before we start: you control readability; the engines control everything after that. Nobody can promise you rankings or recommendations, and this guide won't.

How AI assistants read a store

Assistants get store knowledge two ways: crawlers that index pages ahead of time (OpenAI's GPTBot, Anthropic's ClaudeBot, PerplexityBot and others), and live fetches at answer time, when the assistant opens your page like a fast, impatient reader. In both modes the assistant is doing the same thing: extracting facts from whatever machine-readable material you expose. The mapping is concrete:

Every row is a place where missing or contradictory data silently removes you from the answer. The checklist below works through them, foundation first.

The seven-step checklist

1. Let AI crawlers in

What it is: your robots.txt decides which crawlers may read your store at all. Why agents need it: a blocked crawler indexes nothing; to that assistant your store barely exists. How to check: open yourstore.com/robots.txt and look for Disallow rules under AI user agents such as GPTBot, ClaudeBot or PerplexityBot; also make sure your store isn't password-protected — a storefront behind a password gate is unreadable to every crawler regardless of robots.txt. How to fix: Shopify's default robots.txt allows crawling; if you (or an app) customized robots.txt.liquid to block AI agents, remove those rules deliberately or accept the trade-off knowingly.

2. Structured data that matches the page

What it is: JSON-LD — machine-readable Product and Organization markup embedded in your pages. Why agents need it: it is the difference between an agent parsing "maybe that number near the button is the price" and reading "price": "49.00", "priceCurrency": "EUR". Completeness matters (name, price, currency, availability) and consistency matters more: an agent that finds €49 in the schema and €59 on the page has learned it cannot trust your store, which is worse than learning nothing. How to check: view a product page's source and search for application/ld+json, or paste the URL into Schema.org's validator; compare every schema price and availability against what the page shows. How to fix: most Shopify themes emit basic Product JSON-LD; gaps and mismatches usually come from theme customizations or apps overriding each other. Fix the theme, or use an app-managed schema embed — Sichta's is applied through Shopify's APIs and reversible for 30 days.

3. Complete product data

What it is: the unglamorous admin fields — category, product type, identifiers (GTIN/barcode), named variants, real descriptions. Why agents need it: empty fields are invisible products. An agent matching "organic cotton crew-neck, size M" needs the category, the variant structure and the attributes to exist as data; identifiers let it match your product against reviews and price comparisons elsewhere. Descriptions matter in words, not pixels: an agent cannot read a size chart that only exists as a JPEG. A useful floor: at least a couple of hundred characters of actual descriptive text per product, with key attributes in text or a table. How to check: in Shopify admin, filter products by missing category, type and barcode and count the blanks. How to fix: fill them — tedious, mechanical work. Sichta's catalog gap table lists every product failing each check, exportable as CSV, so it becomes a to-do list instead of a hunt.

4. Alt text on product images

What it is: the text description attached to each image. Why agents need it: for a text-first agent, an image without alt text is a hole in the page — and the same hole affects screen-reader users and image search. Good alt text describes the product factually ("olive green linen duvet cover, buttoned, on a bed"), not promotionally. How to check: in admin, open a product's media and look at each image's alt field; on the storefront, inspect the <img> tags. How to fix: write it per image in admin, or let Sichta draft and apply alt text in one click — with an undo window, and re-verified on the next scan.

5. Policies an agent can quote

What it is: shipping, returns and contact pages with real substance, at Shopify's standard /policies/… URLs. Why agents need it: before recommending a store, assistants check whether the purchase can go wrong: what does shipping cost, how long is the return window, who is the seller. A missing refund policy is a silent disqualifier; a shipping page that never mentions regions or costs answers nothing. If you sell from Germany, Austria or Switzerland, a proper Impressum is also a legal requirement. How to check: read your own policies as a stranger and try to answer "what does delivery to my country cost?" and "how many days do I have to return this?" — with numbers. Then check consistency: if the returns page says 14 days and the FAQ says 30, an agent finds both and trusts neither. How to fix: Shopify's policy settings give you the pages; the content is on you. State regions, costs, times and windows as concrete numbers.

6. Publish llms.txt (and agents.md)

What it is: a curated Markdown index of your store — what you sell, key collections, where the policies live — served at /llms.txt; agents.md is its companion with explicit notes for AI agents. Why agents need it: it is the one page designed for them — small, dense and current — instead of theme markup they must excavate. Honest caveat: an emerging convention that costs little, not a guaranteed channel. How to check: open yourstore.com/llms.txt; a 404 means you don't have one. How to fix: write it by hand and add a URL redirect, or have Sichta generate both files from your live catalog and keep them current. Full detail, including a complete example file, in our llms.txt guide for Shopify.

7. Test answerability

What it is: the end-to-end test — can a buyer's actual questions be answered from your product data alone? Think of the ten questions a buyer puts to an agent: what does it cost, is it in stock, what variants exist, what are the key attributes, does it ship to me, can I return it and within how many days, who is the seller, does it have identifiers, what is it in plain words, and does all of that agree with itself. Why it matters: each unanswerable question is a moment where the agent either guesses, hedges — or recommends a store that does answer. How to check manually: paste a product URL into an assistant and ask those questions; watch where it hedges. How to fix: the fix is always upstream, in steps 1–6. Sichta runs this as a deterministic answerability score — all ten questions against a sample of your products, scored per product and as a store average, alongside 24 checks across crawl access, structured data, catalog completeness and policies.

What you control — and what you don't

Everything above is measurable and within your reach: a file resolves or it doesn't, a schema price matches the page or it doesn't, a return window is stated in days or it isn't. What happens next is not in your control. The engines choose what to crawl, what to trust and what to recommend, and their behavior shifts without notice. That is why the honest goal is readability, not rankings: make your store the one where every fact an assistant needs is present, machine-readable and self-consistent — then let the engines do whatever they do, knowing that when they look at your store, they find answers. For where this is all heading, read our companion piece on agentic commerce readiness in 2026, or run Sichta's free scan to see all seven areas checked against your store in about a minute.