Should I list every product in my llms.txt?

No. A catalog of 4,000 SKUs produces a file nobody can maintain and no client wants to read in one fetch. List the category and collection pages that route to your inventory, a handful of hero products, and the buying guides that answer pre-purchase questions. Product-level detail belongs in a product feed and in schema.org Product markup on each page, where price and stock can update independently of your index. Target takes this approach literally: its file documents the /p/ product URL pattern once instead of listing products.

Does llms.txt improve how my products appear in ChatGPT or Google AI Mode?

Not directly. Conversational shopping surfaces are driven by product feeds and commerce protocols — the Agentic Commerce Protocol published by OpenAI and Stripe, and Google's Universal Commerce Protocol, announced on January 11, 2026 with Shopify, Target, Walmart, Etsy and Wayfair, where a Merchant Center feed with the native_commerce attribute and a profile at /.well-known/ucp do the work. llms.txt is the layer that shapes what an assistant says about your brand, policies and guides. Run both, sourced from the same product data.

Should prices and stock levels appear in llms.txt?

No — link them. Prices and availability change hourly on a sale weekend, and a file that quotes dollar amounts or inventory counts is wrong within days while still being quotable verbatim by an assistant. Keep the file to stable facts: what you sell, where the category pages are, what your returns window is stated as a rule rather than a number you must sync. If you need machine-readable price and availability, publish them in your feed and in Offer markup on the page itself.

E-commerce · October 2026

llms.txt for E-commerce: What an Online Store Should List

A store is the one site type where llms.txt is easy to get badly wrong: the instinct is to export the catalog. Here is the page set that helps an assistant answer questions about your shop, drawn from the files large retailers publish today.

Two Kinds of AI Traffic Hit a Store

A shopping query can end in two very different places. In the first, a person asks an assistant a question about your business — your return window, shipping destinations, which size fits a 26cm foot — and the answer is read off your public pages. In the second, an agent searches a product feed, compares offers and hands the buyer a link or a checkout. Only the first is llms.txt work; the second is feed work, and mixing them up is why so many store files become a 20,000-line SKU dump.

What Large Retailers Actually Publish

Four store files fetched from this server on October 1, 2026, with very different instincts:

Site	Size	What the file is
Target.com	6.2 KB	Category routing, services and fulfillment, owned brands, plus two Optional sections for seasonal pages
Shopify.com	12.5 KB	A twelve-section merchant-facing index: start a business, sell online, payments, sell internationally, plans, developers
Etsy.com	26.8 KB	An instruction manual for assistants: what may be stated confidently, what must not be asserted, buyer and seller routing, then reference lists
Zalando.com	48.4 KB	The largest of the four, at the upper end of comfortable
IKEA.com	404	No file at the root path that day

Two details in Target's file are worth copying regardless of your size. It opens with blockquote notes that document its URL formats — product pages follow /p/{product slug}/-/A-{TCIN}, category pages /c/{category name}/-/N-{node ID} — so a client can construct or recognize URLs instead of guessing. And it ends with two sections marked Optional, the spec's term for content a client may skip under a tight context budget, which is where seasonal and inspiration pages belong.

The Page Buckets Every Store Needs

Skip the catalog export and group into six buckets, in this order:

What you sell. Top-level category pages, one link each — your routing layer and the most-used section.
Buying guides and size help. Fit guides, material comparisons, compatibility tables, which-model-for-X posts.
Policies, stated as rules. Shipping destinations and timeframes as ranges, returns window, warranty, exchanges, price-matching. These are the most quoted lines in any retail file.
Fulfillment and service. Delivery options, pickup and same-day, order tracking, contact and support hours — Target devotes a section to this.
Trust pages. About, ownership or manufacturer status, sustainability claims, accessibility statement, certifications.
Optional. Seasonal campaigns, gift guides, editorial inspiration. Anything with a shelf life.

If your store runs on Shopify, the platform already publishes a generic file for you and the theme override walkthrough covers replacing it.

Product Pages: Patterns, Not Inventory

The question every store owner asks first is whether the file should list products. For a catalog of any size the answer is a pattern plus a shortlist: document how your product URLs are built the way Target does, link the category pages that expose the inventory, and include a curated handful of hero products — not the other 3,900 rows.

The reason is maintenance, not size limits. Product URLs get retired, renamed and re-slugged far more often than policy pages, and a dead link in a curated index costs you the fetch. Page-level detail already has a home: Product and Offer markup on the page itself, which is how the two-file division of labour in llms.txt and Schema.org plays out for stores.

The Sections That Rot: Price, Stock, Promotions

Never write prices, stock counts or promo end dates into the file. A sale weekend changes all three within hours, and the file is the one surface where an assistant may repeat the number as fact. Link the pricing or category page and let it be the source of truth.

Everything else should carry an implicit expiry date. Ask of each line: if this is quoted back to a customer in six months, is it still true? Shipping times and return windows fail that test faster than anything else on a store — the maintenance pass exists for exactly this. Re-check policy links the day after you change a policy, not at the next quarterly review.

Where llms.txt Fits Beside Feeds and Agentic Checkout

Retail is the vertical where an adjacent standard matters most, and llms.txt is not it. The Agentic Commerce Protocol, published by OpenAI and Stripe under Apache 2.0, defines the product feed and checkout exchange that conversational commerce runs on; Google's Universal Commerce Protocol, announced on January 11, 2026 with Shopify, Target, Walmart, Etsy and Wayfair among the co-developers, standardizes discovery, checkout and post-purchase support between agents and merchants. Google's Merchant Center help documentation ties the "Buy" surface to listings carrying the native_commerce attribute.

Those are feed-level, per-SKU, machine-to-machine layers. llms.txt is the human-readable index that shapes the sentence an assistant writes about your brand: what you sell, where your guides are, what your policies say. A store that only ships a feed gets compared on price and availability; a store that also ships an index gets described correctly. The plumbing that connects the file to your crawlers is covered in llms.txt discovery.

Verify It in Three Commands

# status, content type and whether your CDN rewrote the path
curl -sI https://example.com/llms.txt | grep -iE "^(HTTP|content-type|location)"

# size sanity — a store index belongs in single-digit KB, not megabytes
curl -s https://example.com/llms.txt | wc -c

# catch dead category and policy links before an assistant does
curl -s https://example.com/llms.txt | grep -oE "https://[^)]+" | while read u; do
  printf "%s %s\n" "$(curl -s -o /dev/null -w '%{http_code}' -L "$u")" "$u"
done | grep -v "^200"

Then paste the file into the llms.txt checker to confirm the structure held, or build a first draft from your sitemap with the generator and delete the product and tag noise it cannot judge. A store file that hands an assistant the categories, guides and policies in one fetch has done its job.

Files referenced were fetched on October 1, 2026: Target.com 6,158 bytes, Shopify.com 12,520 bytes, Etsy.com 26,822 bytes, Zalando.com 48,385 bytes. Sizes change as these retailers edit their indexes, and a store behind a bot challenge may answer the path with an HTML interstitial rather than Markdown.