llms.txt Examples: What 20 Ecommerce Brands Publish

What llms.txt Is, In One Paragraph

The reason ecommerce brands care: an assistant answering "what does this brand sell and what are its policies" has to assemble that from a themed storefront, and a clean summary file is the cheapest way to hand it the answer. The reason to stay skeptical is in section 11.

The 20 Files, At A Glance

Two brands we tried did not serve a file (one returned an error, one a blocked response), and a few others serve only an HTML 404. Of the twelve DTC brands that responded, ten are on the Shopify template. That is the headline finding, and it needs its own section.

Pattern 1: The Platform-Generated Shopify File

As of September 15, 2026, every Shopify store we checked serves an llms.txt titled "Agent Instructions" followed by the store name, and the file says it mirrors a companion at /agents.md. The brand did not write it; Shopify generates it. We confirmed this the direct way: Bear Basics, our own store, has one, and we never created it.

Read it as an agent would. It is a superb machine-facing document about how to transact with a Shopify store, and it is the same document on every store. An assistant that fetches it learns that Bear Basics is a Shopify store with a refund policy. It learns nothing that would help it recommend a tallow balm to someone with sensitive skin. Nine of the ten DTC brands on the template, some of them among the most recognized in their categories, are in the same position.

That is not a criticism of Shopify; the file is doing agentic-commerce plumbing and doing it well. The UCP vs ACP guide covers what those endpoints mean for a brand. It is a warning to Shopify brands: you have an llms.txt, it is not about you, and the question is whether you can add to it or replace it. Section 12 covers the options.

Pattern 2: The Curated Index (Warby Parker)

Warby Parker's file is 1.7 KB and it is the best DTC example we found, because it does three things most files do not.

• It points to Markdown mirrors of key pages (/llms/pages/get-a-prescription.md, /llms/pages/history.md and so on), so the assistant gets clean text instead of a themed page, and each link carries a one-line summary that says what the page is for.
• It states where the source of truth is for volatile facts. The file tells the model to use the live HTML pages for pricing, promotions, appointment availability, store hours, insurance and product availability. That single paragraph prevents the most common assistant error about a retailer: quoting a stale price.
• It groups by intent: vision care and prescriptions, company information, education. Not by site navigation.

The whole file would take an afternoon to build for most brands. It is the pattern our template in section 10 is based on.

Pattern 3: The Full Catalog (Nomad)

Nomad's file is 109 KB. It opens with a strong one-paragraph description (what they make, the materials, the devices), then lists collections with summaries, support, content and policies, and then 473 product lines, each a linked product name with the first sentence of its description.

The case for it: an assistant that reads it knows the whole catalog, including which products exist for which iPhone, without crawling 473 pages. The case against: at 109 KB it is well beyond what most retrieval systems will read in one pass, product descriptions go stale as the catalog changes, and a full catalog belongs in a product feed or llms-full.txt rather than the index file the spec describes. Nomad's top section, on its own, would be a near-perfect 5 KB file.

If you sell fewer than about fifty products, the full-catalog pattern is workable and the descriptions are the valuable part. Above that, split it: the index in llms.txt, the catalog in llms-full.txt or a linked feed, and keep the index under 10 KB.

Pattern 4: The Platform Index (Shopify, BigCommerce, Klaviyo, Gorgias)

The commerce platforms and the SaaS tools ecommerce brands run on all publish curated indexes, and they are worth reading because they are written by teams with documentation budgets.

Klaviyo's is the most instructive file in the set, because it treats llms.txt as instructions rather than a sitemap. It tells the model which source wins when two pages disagree, how the product names relate to each other, and when it was last checked. A brand's version of that is three lines: use the product page for price and stock, use the FAQ for policy, and here is the date.

What The SaaS Files Do Better

Stripe, Cloudflare, Zapier and Vercel are not ecommerce brands, and their files are in this list because they show conventions that DTC files have not adopted yet.

• Separate files per audience. Stripe serves a 65 KB marketing file at stripe.com and a 91 KB developer file at docs.stripe.com. A brand's equivalent is a customer-facing file at the store and, if you have a wholesale or affiliate site, a separate one there.
• Stable anchors. Zapier's file opens by saying its section anchors are stable and safe to deep-link (llms.txt#mcp, for example). A brand can do the same: #products, #policies, #ingredients, and never rename them.
• A companion full file. Cloudflare and others link an llms-full.txt for the complete content. That is where Nomad's 473 products belong.
• Summaries that say what the page answers, in Vercel's case in one clause each. The test for a summary line is whether a model could decide to fetch the page from the summary alone.

Our Own File, As An Example

evolveamz.com/llms.txt is 10 KB, hand-written, and organized by the clusters we most want assistants to cite: an About section, the AI search resource hub as priority content with its guides grouped into definitions, comparisons and decision frameworks, then services, then a link to the blog and the sitemap. The blockquote states who we are for in one sentence, with the founding year and location.

What we would change, having read the other nineteen: add a last-updated line at the top (Klaviyo), add a source-of-truth paragraph pointing pricing questions to /pricing/ (Warby Parker), and add stable anchors (Zapier). We also found two lines in the file that use terminology we retired earlier this year, which is the other lesson: a hand-written file goes stale like any page, and it needs an owner and a quarterly read. Bear Basics is the mirror case: a platform file with no brand content, which we will add to as section 12 describes.

What Every Good File Includes, And What They Skip

The other file most brands need alongside this one is structured data; the schema markup stack guide covers Product, Organization and FAQ schema, which are the machine-readable layer assistants are confirmed to use.

The Template

Built from Warby Parker's structure, Klaviyo's rules and Nomad's blockquote. Replace the bracketed lines; delete sections you do not have.

Target size: 3 to 8 KB. Summaries are facts. Anchors (the H2 names) never change once published. The file is plain text served at /llms.txt with a text/plain or text/markdown content type; a WordPress site can serve it from the web root, and the Shopify case is in section 12.

Does Any Assistant Read It?

Here is the honest state of the evidence, since we sell this work. llms.txt is a proposal. As of this writing, no major AI assistant operator has publicly confirmed that its retrieval or crawling systems read llms.txt or use it to decide what to cite, and at least one search engine representative has said publicly that their systems do not use it. Adoption is strongest among developer-documentation sites, where the files are read by coding tools that were built to look for them, and the Shopify template is aimed at shopping agents rather than chat assistants.

So the case for publishing one is modest and specific. It costs an hour. It is a clean, crawlable, plain-text summary of your site, which is useful to any system that fetches it and harmful to none. If an operator does start reading it, the brands with a good file are ahead by a year. And the crawler log method covered earlier in this series will tell you whether anything is actually fetching yours, which is the only measurement that matters.

What it is not: a substitute for schema, for a crawlable site, for content worth citing, or for the listings and reviews that assistants demonstrably use. The AI search visibility guide ranks those; llms.txt is near the bottom of the list on evidence and near the top on cost-per-hour, which is the reason to do it and the reason not to expect much.

Publish It In An Hour

The Shopify step is the unresolved one, and we are not going to pretend otherwise: the platform owns the path, the file it serves is useful for agents and empty about your brand, and whether you can override it depends on your setup and on Shopify's current behavior, which is changing quickly as agentic commerce rolls out. Check the documentation on the day you do this. Everyone else can have a good file live before lunch.