More of your customers are asking an AI before they ask you. They have ChatGPT compare two of your products, get Perplexity to summarize your return policy, read a shopping assistant’s pick and never open a single tab. When that happens, a machine reads your store on your behalf. The question is whether it reads the right things.
That’s the small idea behind llms.txt. It’s a plain text file you publish at the root of your domain, written in markdown, that gives large language models a clean, curated map of what matters on your site. Think of it as a sitemap, but written for a reader with a short attention span and a strong preference for plain language.
It’s worth understanding properly before you rush to add one. Some of what’s said about llms.txt is real. Some of it is wishful. Let’s separate the two, then get into how you actually ship one on Shopify, which is less obvious than it sounds.
What llms.txt actually is
The file lives at yourstore.com/llms.txt. It’s markdown, not code, and the format is deliberately simple. One H1 with your store or brand name. An optional short summary in a blockquote. Then a few sections, each a list of links with a sentence explaining what’s on the other end.
A trimmed example for a store might read like this:
# Northbound Coffee
> Single-origin coffee, roasted to order.
## Shop
- [Subscriptions](https://...): set your cadence
- [Single bags](https://...): the lineup + notes
## Help
- [Shipping & returns](https://...): how it ships
- [Brew guides](https://...): ratios per methodThe reasoning behind it is genuinely sensible. A language model works inside a limited context window, so it can’t read your whole site at once. Your actual pages are buried in navigation, scripts, banners, popups, and tracking. That’s noisy and expensive for a machine to parse. A short markdown file strips all of that away and says, in effect, here’s who we are and here are the pages worth reading. Jeremy Howard, who proposed the idea in late 2024, framed it as helping models find the signal without wading through the HTML.
So far, so reasonable. Now the part most articles skip.
What it does not do
llms.txt is a proposed convention, not an adopted standard. That distinction matters, and glossing over it is how stores end up disappointed.
As of now, the big AI companies have not committed to reading it. Google has said plainly that its systems don’t use llms.txt for search or AI overviews. The major crawlers haven’t announced support either. There’s no guarantee that publishing the file changes what ChatGPT or any assistant says about you tomorrow. Anyone telling you it’s a ranking factor is ahead of the evidence.
Here’s the honest version. Right now llms.txt is closer to a bet than a lever. It’s a low-cost, well-formed signal that some tools already read and more might. It is not a switch that makes the AIs cite you. If you came looking for a trick that vaults you to the top of an AI answer, this isn’t it, and neither is anything else. That game is still won the slow way, with clear product pages, clean structured data, honest specs, and a name other sites mention. We wrote about that in getting cited by ChatGPT, and nothing about llms.txt replaces it.
So why bother
Because the downside is tiny and the upside is real, and that’s a trade worth taking.
The file costs you an afternoon and a few kilobytes. There’s no risk to your SEO, no effect on customers, who’ll never see it. If adoption grows the way robots.txt and sitemaps once did, the stores that already have a clean, accurate file are ready, and the ones that don’t are scrambling. You’re not betting the business. You’re buying an option for almost nothing.
There’s a quieter benefit too. Writing a good llms.txt forces you to answer a useful question. If I had thirty seconds to tell a stranger what my store is and which ten pages matter most, what would I say? Most store owners have never written that down. The exercise tends to surface pages that are doing real work and pages that are just taking up space.
And it gives you a sliver of control. Left alone, an AI forms its impression of you from whatever it happens to scrape, which might be an old blog post or a thin collection page. llms.txt lets you put your best, most current pages in front of it on purpose. It’s not a guarantee the model listens. It’s the difference between handing someone a menu and letting them guess from the smell.
The Shopify catch
Here’s where Shopify makes you work for it. On most platforms you drop a file in a folder and you’re done. Shopify owns your URL structure, and it doesn’t hand you the root. You can’t just upload a file and have it appear at yourstore.com/llms.txt with the right content type. The paths are fixed: /products, /collections, /pages, and so on.
You can edit robots.txt through robots.txt.liquid, and you can upload files in the admin, but neither gets you a clean markdown file served from the exact root path that the convention asks for. So the question on Shopify isn’t really “what do I write.” It’s “how do I serve it.” A few routes actually work, with honest trade-offs on each.
A purpose-built app. Several apps in the Shopify App Store now generate and serve an llms.txt for you, usually by building the file from your products and pages and keeping it current as your catalog changes. This is the least technical path. The cost is another app on your store, which longtime readers know we’re cautious about, so weigh whether one more recurring subscription and one more script earns its place.
An edge proxy. If your store sits behind a CDN like Cloudflare, you can serve /llms.txt from the edge, before the request ever reaches Shopify. A small worker returns your markdown with the correct text/plain content type, at the real root, with nothing added to the store itself. This is the cleanest result and the one we usually reach for. It does assume you’ve got that layer in place and someone comfortable configuring it.
A page, with caveats. You can create a Shopify page and put the content there, then point a URL redirect from /llms.txt to it. Be clear-eyed about what this gives you. The redirect serves HTML wrapped in your theme, not a raw markdown file, and the URL changes on the way. It’s the weakest option, a partial fix at best, but it’s better than nothing if the other two are out of reach.
Whichever route you pick, the file itself needs to be right, and it won’t stay right on its own.
Writing one that’s actually good
A bad llms.txt is worse than none, because it tells the machine something false with total confidence. A few things keep yours honest.
Keep it short and curated. The point is signal. Link the pages that matter, your key collections, your best products, your policies, a few strong guides, and leave the rest out. A file that lists all four hundred products is just your sitemap again, and it buries the things you actually want read.
Write the descriptions for a person. Each link gets a plain sentence about what’s there. Skip the marketing fog. “Subscriptions: pick a roast, set a cadence, skip anytime” beats “Discover our revolutionary subscription experience,” and it’s the kind of clear, liftable line a model can quote without mangling.
Tell the truth, exactly. This file may become part of how an AI describes you to a customer. If your return window is 30 days, say 30, not “hassle-free returns.” Wrong details here don’t just fail to help. They actively mislead a machine that’s speaking for you.
Keep it current. The single fastest way to make llms.txt useless is to write it once and forget it. Products get discontinued, policies change, links rot. A stale file teaches the AI yesterday’s catalog. If you go the app or proxy route, wire it to regenerate as the store changes. If you maintain it by hand, put a recurring reminder on it like any other piece of living content.
The honest bottom line
llms.txt is a small, sensible idea with an uncertain future. It might quietly become standard, the way sitemaps did. It might fade. Nobody can tell you which yet, and you should be suspicious of anyone who claims to know.
What we can tell you is the math. The cost is an afternoon. The risk is zero. The upside, if AI shopping keeps growing the way it has, is that your store already speaks the machines’ language while your competitors are still figuring out the question. That’s a cheap bet on a likely future, and we’ll take that trade most days.
Just don’t mistake it for the whole job. llms.txt points the AI at your best pages. Whether those pages are worth quoting is still on you.
PixelSplash designs, builds, and scales Shopify storefronts for brands that have outgrown their theme. Want a clean llms.txt served from your real root, plus the product pages and structured data to back it up? Book a free consult. 30 minutes, zero pitch.