Skip to content
E-Commerce Schema

How to Build /llms.txt and Grounded Product JSON-LD for AI Shopping Assistants

A practical blueprint for an 8 KB /llms.txt and Product JSON-LD where every value matches the visible page.

BuyerCite Editorial & GEO Research Published Updated 4 min read · 864 words

Key Takeaways

  • /llms.txt is a Markdown manifest of your best sources; it never replaces robots.txt.
  • Keep it small: summary, key guides, catalog entry points and policies.
  • Every Product and Offer value must match the visible page right now.
  • Use ProductGroup for variants and monitor price and availability parity continuously.
On this page
  1. What /llms.txt is — and what it is not
  2. Anatomy of a useful /llms.txt for a store
  3. Keep a strict size budget
  4. Write link descriptions as facts
  5. Grounded Product JSON-LD
  6. Grounding rules we enforce
  7. Variants with ProductGroup
  8. Validation and monitoring
  9. How /llms.txt, schema and the sitemap fit together
  10. Common /llms.txt mistakes

AI shopping assistants answer questions like "which 60-quart mixer ships fastest under $9,000?" by combining crawled pages, structured data and live fetches. Two artefacts make your catalog easier to ground: an /llms.txt manifest that points models at your best sources, and Product JSON-LD whose every value is true on the visible page.

What /llms.txt is — and what it is not

/llms.txt is a proposed convention (documented at llmstxt.org): a Markdown file at the root of your site with an H1, a one-paragraph summary in a blockquote, and sections of links to the pages a language model should read first.

Be precise about its limits:

  • It is not an access-control file. Crawlers still obey robots.txt, not /llms.txt.
  • Support varies by vendor and changes over time; treat it as a low-cost, high-clarity hint rather than a ranking factor.
  • It only helps if the linked pages are themselves fetchable and extractable.

Anatomy of a useful /llms.txt for a store

# Northwind Supply

> Northwind Supply sells rebuilt commercial bakery equipment — spiral
> mixers, sheeters and deck ovens — with a 12-month parts warranty and
> nationwide freight from Skokie, Illinois.

## Buying guides
- [Choosing a spiral mixer size](https://northwind-supply.example/guides/mixer-size): capacity by daily dough volume
- [Rebuilt vs new mixers](https://northwind-supply.example/guides/rebuilt-vs-new): warranty and lifespan comparison

## Catalog
- [Spiral mixers](https://northwind-supply.example/collections/spiral-mixers): 20–120 qt, in stock
- [Shipping & freight](https://northwind-supply.example/shipping): lead times by region

## Policies
- [Warranty](https://northwind-supply.example/warranty): 12 months parts, 90 days labour
- [Returns](https://northwind-supply.example/returns): 14-day return window on uninstalled units

Keep a strict size budget

Write for a context window, not for a sitemap. We cap our own /llms.txt at 8 KB: the summary, the dozen pages that answer most buyer questions, and the policies a model must not get wrong. Link to collections and guides rather than enumerating thousands of SKUs.

"Choosing a spiral mixer size: capacity by daily dough volume" tells a model what it will find. "Learn more" tells it nothing.

Grounded Product JSON-LD

The rule is simple: every value in the markup must be visible on the page and true right now. Here is a complete, grounded Product node:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "@id": "https://northwind-supply.example/products/m60#product",
  "name": "Northwind M60 Rebuilt Spiral Mixer, 60 qt",
  "sku": "NW-M60-R",
  "gtin13": "0000000000000",
  "brand": {"@id": "https://northwind-supply.example/#organization"},
  "description": "Rebuilt 60-quart spiral mixer, 2 HP, 208V three-phase.",
  "image": ["https://northwind-supply.example/media/m60-front.webp"],
  "itemCondition": "https://schema.org/RefurbishedCondition",
  "offers": {
    "@type": "Offer",
    "url": "https://northwind-supply.example/products/m60",
    "price": "8450.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock",
    "priceValidUntil": "2026-12-31",
    "hasMerchantReturnPolicy": {
      "@type": "MerchantReturnPolicy",
      "applicableCountry": "US",
      "merchantReturnDays": 14,
      "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow"
    }
  }
}

Grounding rules we enforce

  1. Price parity. The price string equals the price rendered in the buy box, after currency formatting is removed.
  2. Availability parity. InStock only when the add-to-cart button is enabled.
  3. Condition honesty. Rebuilt or used goods use RefurbishedCondition or UsedCondition, never NewCondition.
  4. No invented ratings. Omit aggregateRating unless real, on-page reviews exist.
  5. Freshness. Record when each price was observed and regenerate markup when the catalog changes.

Variants with ProductGroup

When one page sells several sizes or voltages, describe the family once and each purchasable variant separately:

{
  "@type": "ProductGroup",
  "name": "Northwind M-Series Spiral Mixers",
  "productGroupID": "NW-M",
  "variesBy": ["https://schema.org/size"],
  "hasVariant": [
    {"@type": "Product", "sku": "NW-M40-R", "size": "40 qt",
     "offers": {"@type": "Offer", "price": "6900.00", "priceCurrency": "USD"}},
    {"@type": "Product", "sku": "NW-M60-R", "size": "60 qt",
     "offers": {"@type": "Offer", "price": "8450.00", "priceCurrency": "USD"}}
  ]
}

This lets an assistant answer "how much is the 40-quart version?" from the markup instead of guessing from prose.

Validation and monitoring

  • Validate syntax with the Schema Markup Validator and eligibility with Google's Rich Results Test.
  • Diff rendered price and availability against the JSON-LD in CI or on a nightly crawl; alert on any mismatch.
  • Re-fetch /llms.txt after each deploy and check the size stays under budget and every link returns 200.
  • Poll shopping prompts in AI answer engines and record whether answers quote your current price — the end-to-end test that matters.

How /llms.txt, schema and the sitemap fit together

Each artefact answers a different question for a machine reader:

Artefact Question it answers Audience
robots.txt May I fetch this URL? Every crawler
sitemap.xml Which URLs exist and when did they change? Search and AI index crawlers
/llms.txt Which pages should I read first, and what do they contain? Language models and agents
Product JSON-LD What exactly is this item, its price and its availability? Search engines and shopping assistants

Keep them consistent. A URL listed in /llms.txt must be allowed in robots.txt, present in the sitemap and return 200. A product price must be identical in the buy box, the JSON-LD and any feed you submit to merchant platforms.

Common /llms.txt mistakes

  • Dumping the sitemap. Thousands of links defeat the purpose of a curated manifest.
  • Marketing copy instead of facts. "World-class quality" gives a model nothing to quote; "12-month parts warranty" does.
  • Stale links. Redirect chains and 404s in the manifest waste a model's fetch budget and erode trust.
  • Listing blocked pages. If a page is disallowed for AI search crawlers, linking it in /llms.txt only creates confusion.
  • Forgetting policies. Shipping, returns and warranty pages are where assistants most often get facts wrong; link them explicitly.

Frequently Asked Questions

Do AI crawlers obey llms.txt instead of robots.txt?

No. robots.txt remains the access-control file; llms.txt is an optional guide to your most useful pages and its support varies by vendor.

Should every SKU be listed in llms.txt?

No. Link to collections, buying guides and policies, and keep the file within a small size budget so it fits comfortably in a model's context.

Can I add aggregateRating without on-page reviews?

No. Only mark up ratings that are visible on the page and come from real customer reviews.

Written by
BuyerCite Editorial & GEO Research
AI Search & Commercial Citation Engineering

The BuyerCite research desk builds and audits the platform's AI answer tracking across ChatGPT Search, Perplexity, Google AI Overviews, Google AI Mode, Gemini and Claude, and publishes the methods it uses to measure citations, brand naming and AI crawler behaviour.

Published by BuyerCite, operated by DIGISOIL LIMITED (Company No. 09410315).

  • llms.txt
  • Product Schema
  • JSON-LD
  • E-Commerce