llms.txt for Shopify: What I Put in Ours, Who Reads It

llms.txt Shopify, from someone who hand-wrote one — what went in ours, what a store's file needs, and honest evidence on whether AI ever reads it.

Linh Nguyen · Updated 2026-07-04

The week we launched storefleet.io.vn in June 2026, one of the last files I wrote by hand was llms.txt — a plain Markdown file sitting at our domain root, addressed to a reader that may not exist. Not Google. Not a human. An AI assistant that might, someday, want to know what our product is without parsing our whole site. It took me about twenty minutes (my estimate; nobody was timing), and I still cannot prove a single machine has ever read it with intent.

That tension is the whole subject of this post. I run five Shopify stores alongside our own site, so I've faced the "llms.txt Shopify" question both as a SaaS founder and as a store operator. Here's what the file really is, what I put in ours and left out, what a store's version should contain, and the honest evidence on whether anyone reads these files at all.

A Markdown file addressed to machines: where llms.txt actually comes from

First, the origin, because it gets misattributed constantly: llms.txt is not a Google thing, not an OpenAI thing, and not a standard. It's a proposal published in September 2024 by Jeremy Howard of Answer.AI, and the spec is almost disappointingly simple: an H1 with your name, a blockquote summarizing what you are, then sections of Markdown links with one-line descriptions of your most important pages.

The comparison everyone reaches for is robots.txt, but the two files point in opposite directions. Robots.txt is a bouncer — it tells crawlers what they may not touch. An llms.txt is a menu: here's who we are, here's what matters, start here. The reasoning behind it is sound even if adoption isn't: language models work inside limited context windows, and a modern web page buries its actual content under navigation, cookie banners, and JavaScript. A curated plain-text summary skips all of that.

Sound reasoning, though, is not the same as evidence that anyone consumes it — hold that thought for two sections.

What went into ours — and the first draft I threw away

My first draft was bad in an instructive way: I essentially pasted our homepage copy into a text file. Reading it back, it was ad copy — superlatives, feature language, the voice you use when persuading a human with eyes. An AI assistant summarizing "what is this site" doesn't need persuading; it needs facts it can repeat without embarrassing anyone. I threw that draft away and rewrote it the way I'd brief a new hire.

What the final file contains: one plain sentence on what the product is (a multi-store dashboard for Shopify operators), links to the handful of pages that answer real questions — features, pricing, the guides we most want quoted accurately — and how to reach us. What I deliberately left out: the full blog archive (a link dump helps no one), anything with a superlative in it, and anything that changes weekly. The file regenerates as part of our site deploy, so it can't silently drift out of date — a stale llms.txt feeding an AI last quarter's pricing would be worse than no file at all.

One humbling footnote: weeks after I wrote it, I discovered our CDN had been silently blocking every AI crawler the file was meant to welcome — the full Cloudflare story is in our AI SEO vs GEO write-up. Whatever you ship, verify a bot can actually fetch it.

If I wrote one for my stores: what an agent would actually need

Writing the file for a SaaS site made me think about what the equivalent looks like for a store, because it's not the same document. Product pages already tell an AI what you sell. What they bury are the operational answers an assistant needs before it can recommend you: how long shipping takes and to where, what the returns window is, how sizing runs, and how a human gets help when something goes wrong. Those are exactly the questions I'd want answered before an agent put one of my five stores in front of a customer.

So a store's llms.txt, in my view, looks like this:

# Your Store Name

> One plain sentence: what you sell and who it's for.

## Products
- A couple dozen best sellers, one line each — not the catalog

## Policies
- [Shipping](link) — times, zones, costs
- [Returns](link) — window and conditions
- [Sizing](link) — if you sell anything wearable

## Contact
- [Support](link)

Write it in the flat language you'd use answering a support ticket, and if you sell hundreds of products, resist listing them — a short list of what actually drives revenue is enough.

Getting it live on Shopify — check first, you may already have one

Before building anything: visit yourstore.com/llms.txt. Shopify has been rolling out native llms.txt files for stores, generated automatically with store metadata and MCP endpoints. If yours is there, the platform has done the baseline for you — don't pay anyone to "implement" what already exists.

If you want a curated file of your own instead of the auto-generated one, the free path still works: write the file, upload it under Content → Files in the admin, then point a URL redirect from /llms.txt to the CDN URL Shopify gives you. It's minutes of clicking, not an engineering project. The honest caveat: that's a 301 redirect, and there's no guarantee every AI crawler follows redirects. Running several storefronts? This is exactly the kind of small repeated task that argues for managing multiple Shopify stores from one place.

What I would not do is pay a monthly app subscription for this. It's a static text file. Renting one is how you turn a free hedge into a recurring cost.

The part no one selling GEO services will tell you

Now the evidence question, with no hedging: no major AI assistant is confirmed to fetch llms.txt at scale. The largest study I know of — SE Ranking's analysis of 300,000 domains, reported by Search Engine Journal — found no statistically significant relationship between having the file and earning AI citations. Anthropic publishes an llms.txt for its own docs, which tells you the idea has sympathizers, not that any crawler consumes yours.

My own data point is just as flat: since June 2026 I cannot attribute one visit, one citation, one anything to our file. There's no Search Console for ChatGPT, no dashboard, no feedback loop. If AI referral traffic ever shows up for us, I expect to see it first in referrer logs from the citation-heavy engines — Perplexity and Google's AI Mode — not from anything llms.txt did.

Why I ship it anyway

Here's where I land, and it's the position I'd defend to another operator: llms.txt is probably decorative today, and I ship it anyway, because twenty minutes of writing against zero downside is a trade you take even at long odds. What I won't do is pay for it — no apps, no retainers, no "GEO package" where this file is a line item. The moment someone charges recurring money for a static Markdown file, the economics that justified the experiment are gone.

And keep it in proportion: the file is one small piece of making a store legible to AI. Crawler access, structured data, and clean product data do more of the real work, and that whole stack is covered in our guide to optimizing your store for AI discovery. An AI can only represent your store as well as your data allows — which is why the boring work of keeping orders, inventory, and product data clean and in one place beats any single file. The llms.txt is free either way.