llms.txt for an online store: worth it or not?

What an llms.txt is

An llms.txt is a text file, written in Markdown, placed at the root of a site: example.com/llms.txt. Jeremy Howard proposed it on 3 September 2024, on llmstxt.org. The idea: give language models a short summary of the site and a list of links to the pages that matter, instead of letting them guess from the HTML.

The format is simple. A # title with the site's name (the only required part), a > quote that sums it up, then ## sections with lists of links. Its authors say it themselves: the file is mostly useful when a model answers (inference), not when it is trained.

It is a proposal, not a standard. No standards body has adopted it, and no crawler is bound to read it. So the question for a store is simple: who reads it, and what does it change for sales?

The facts in this guide were checked on 9 October 2026, on the pages linked.

What Google says, in writing

Google settled it in its AI optimization guide, updated on 10 July 2026:

"It's completely fine if you decide to create and maintain LLMS.txt files (or other similar files) for other services or systems that use these files. Doing so will neither harm nor help your site's visibility or rankings in Google Search, as Google Search ignores them."

In other words: Google Search ignores these files, and that includes its AI features (AI Overviews, AI Mode). The same guide says no special file is needed to appear in them. For products, it points to what already existed: structured data and Merchant Center feeds.

AI vendors: nothing documented

OpenAI, Anthropic and Perplexity document their crawlers, what each one does and how to block it. None of these pages says their crawlers read a site's llms.txt:

• OpenAI's page on its crawlers;
• Anthropic's;
• Perplexity's.

One detail shows the confusion well: OpenAI and Perplexity publish an llms.txt for their own documentation. That is a real use, handy for developers who plug a coding assistant into technical docs. It says nothing about what ChatGPT or Perplexity do with a store's file.

Who actually reads it

On 15 June 2026, Ahrefs published a study of 137,210 domains, measured in May 2026:

• 28% of these domains publish an llms.txt;
• 97% of those files received no request at all over the month;
• the remaining 3% (about 1,100 domains) got all the traffic measured;
• training crawlers account for only 5.3% of requests, 4.51% of them from GPTBot;
• Claude-Code, Anthropic's coding assistant, fetched it more than any AI search or assistant crawler;
• SEO audit tools make up about a fifth of requests (21.7%).

The study says it plainly: nothing shows that crawlers act on what they fetch, and the file "won't help your AI search visibility today". No crawler comes looking for an llms.txt that does not exist.

The readers are easy to describe: coding assistants and audit tools, not the assistants that recommend products to your customers.

Lighthouse: an informative audit

Lighthouse, Chrome's audit tool, has an "Agentic Browsing" category with an llms.txt audit. It is informative, not scored. With no file (a 404), it is marked "not applicable", since the documentation says the file is optional for now. It only flags a problem when the server returns an error.

If an audit tool takes points off because you have no llms.txt, it is stricter than Google and Chrome.

Shopify already serves one

On 9 October 2026, we opened the llms.txt of four public Shopify stores. Three of them served one they had not written themselves: Brooklinen, Tentree and Allbirds. Their robots.txt opens with comments pointing to /agents.md and to a /.well-known/ucp profile. The llms.txt mirrors the content of agents.md. The fourth store, Gymshark, had none.

This is an observation, not documentation: we found no Shopify page announcing it. The content is telling. It is not a catalog summary. It is a manual for shopping agents: how to search the catalog, build a cart and check out through the UCP protocol, with the buyer approving the payment.

For a Shopify merchant, the takeaway is practical. The file may already be there, and it does not talk about your products.

What a store should do instead

The assistants that recommend products read pages, feeds and third-party sources. That is where visibility is decided, and each point below has a documented effect:

1. robots.txt: let the assistants' search crawlers in, such as OpenAI's OAI-SearchBot. Our robots.txt checker tests it in seconds.
2. Server rendering: the title, price and description must be in the HTML. Most crawlers do not run JavaScript.
3. Structured data Product and Offer, with a price, currency and availability identical to the page.
4. Product feeds: Merchant Center for Google, OpenAI's specification for ChatGPT. The ChatGPT Shopping guide lists the fields.
5. Consistency: one price and one availability everywhere, page, markup, feed and marketplaces.

The e-commerce GEO guide covers these workstreams in order.

If you want to publish one anyway

The cost is low and the risk is nil, as long as the file stays accurate. A few rules:

• short: the store's name, one sentence on what it sells and where it ships, a handful of links;
• stable: links to collections and information pages, not to products that come and go;
• no prices or stock: what changes every week will end up wrong in a file nobody updates.

# Example Linen Co.

> Washed linen bedding, made in Portugal. Ships to the UK and the EU.

## Catalog

- [Duvet covers](https://www.example.com/collections/duvet-covers)
- [Sheets and pillowcases](https://www.example.com/collections/sheets)

## Information

- [Shipping and returns](https://www.example.com/pages/shipping)
- [Size guide](https://www.example.com/pages/size-guide)

On Shopify, check what the store already serves at that address before touching it.

What we do with it

Our site publishes an llms.txt. It costs nothing, it is generated from the same copy as the pages, and we do not count on it to be cited.

In the product, skoup's technical audit notes whether an llms.txt is there. It weighs little: 5 points out of the 100 of the "Crawlability" block, which counts for a quarter of the score. A missing file is a warning, never an error. When a store publishes one, skoup reads it first to understand what the store sells. The points that really matter are the ones in the list above.

To find out whether ChatGPT cites your store today, the free audit asks three real questions and shows the brands recommended instead of yours. The glossary defines the terms used in this guide.