llms.txt: what it does, who reads it, how to do it right
A short guide to llms.txt: what the file is, which tools actually read it, and how to publish a clean one on WordPress and Shopify, with a two-command check.
llms.txt is a short Markdown file at the root of your site that gives AI agents a map of the pages worth reading first: your name, a one-paragraph summary and a list of links with a line each. Today it is read mostly by AI agents, coding assistants and site-audit tools, and only rarely by the bots that answer search questions. It is cheap AI-readiness hygiene, and it only helps if the file is accurate, current and served as plain text.
Most advice about the file stops at "add one". The problems we ran into while generating ours were in the details: titles with phone numbers in them, a file that described the site as it was last spring, a /llms.txt that answered with an HTML page. This post covers what the file is, who requests it, and how to publish one that is right.
What is llms.txt?
It is a proposal by Jeremy Howard, published in September 2024 at llmstxt.org. The only required part is an H1 with the site name. After it come an optional blockquote summary and H2 sections that are lists of links, each with an optional note. A section called Optional marks links an agent can skip when it is short of context.
It is not robots.txt. Robots.txt tells crawlers what they may fetch; llms.txt is meant to be read on demand, when an agent is helping one person, as a map of where to look. An illustrative file:
# Example Linen Shop
> Bed linen made from washed flax, sold online and shipped within the EU.
## Guides
- [Choosing linen weight by season](https://example.com/guides/linen-weight): GSM explained
- [Caring for washed linen](https://example.com/guides/linen-care): washing and creasing
## Optional
- [Shipping and returns](https://example.com/policies/shipping)Who reads it today?
Ahrefs's June 2026 study logged who requested the file across 137,210 domains in May 2026. About 28% of those domains had a valid llms.txt, an upper bound because Ahrefs customers skew technical.
| Who requested llms.txt | Share of all requests |
|---|---|
| SEO audit tools | 21.7% |
| AI retrieval bots | 1.1% |
SEO audit tools made the largest share of requests. Retrieval bots, the ones that fetch pages to answer a user's question, made 1.1%. A coding assistant, Claude Code, fetched the file more often than any retrieval bot or assistant. The realistic readers today are agents, coding assistants and audit tools. Ahrefs also found that 97% of the files in its sample were not requested at all in that month, so publishing one is no guarantee of readers.
What the file does not do on its own is rank you or get you cited. Ahrefs notes that Google's guidance says files like this are not needed for generative AI search, and that a fetch is not a read. Whether your pages are found and quoted still depends on crawl access and quotable first paragraphs; we cover both in what GEO means for a small business site.
What goes in the file
An agent that reads the file takes it at its word, and it is published under your name, so the details matter. What we learned generating ours:
- A summary in plain words: what you sell, to whom, where. One paragraph, not a brochure.
- Ten to thirty links, one line each, with the rest under Optional. A sitemap already lists every page; this file is the short list.
- Names written for people. Title tags are written to win a click, so they carry phone numbers, calls to action and, on page-builder sites, leftover menu labels. Built from the title tags of a real WordPress site, our first directory had a title ending in "Call now", another that was only a menu label in square brackets, and a phone number that did not match the one in the page header. We now strip all of it, and drop any description sentence that contains a phone number or address.
- Only facts you would print on a business card. A phone number or address you have not verified does not belong in a file that goes out under your name.
- Descriptions, not instructions. An agent may act on text it reads here, and Ahrefs advises treating the file like code.
Publishing it on WordPress
The simplest route is a text file named llms.txt in the folder that holds wp-content. Some SEO plugins generate one instead; All in One SEO says its free version builds it from recent posts (a vendor claim).
If a plugin serves the file and /llms.txt returns 404, open Settings, Permalinks and press Save Changes, which rebuilds WordPress's rewrite rules. We reproduced the same failure while building our own plugin: a rewrite flush that ran before the /llms.txt rule existed left the path at 404 for good, and the symptom looked like nothing had ever been published.
Freshness is the other trap. A file written once describes the site as it was. On a test site ours went live listing one article while the current preview held 31 entries, because the update was waiting for a six-hour schedule. It now also refreshes when something is published or approved.
Publishing it on Shopify
Do not upload anything. Since May 2026 Shopify serves a default /agents.md for every store, and /llms.txt and /llms-full.txt point to the same content (Shopify changelog, 28 May 2026). Shopify's docs call it the canonical, agent-facing description of a store: the default covers discovery and MCP endpoints, read-only browsing URLs, published policies and guidance for shopping agents, and leaves out contact details.
To describe your own catalogue, add templates/llms.txt.liquid to the theme for /llms.txt alone, or templates/agents.md.liquid to change all three paths. Merchants in Shopify's developer forum report that URL redirects created in the admin stopped working once this began. One consequence: a check that only asks whether /llms.txt answers now passes every Shopify store, so it tells you nothing about yours.
Check yours in two commands
curl -sI https://yoursite.com/llms.txt
curl -s https://yoursite.com/llms.txt | head -5You want a 200, a content-type of text/plain or text/markdown, and a first line that starts with # . A 200 alone proves little: in Common Crawl's July 2026 experiment, about 40% of the 200 responses to /llms.txt and /llms-full.txt were HTML, usually a single-page app's shell answering a path it did not know, and about 11% were robots.txt content served at those addresses. If a CDN rule blocks AI user agents, the file never reaches them either.
Where Mose fits
On WordPress sites Mose generates /llms.txt from the brand profile you approve and the pages it has read, refreshes it when something is published or approved, and serves nothing rather than an empty file. The full GEO pack on the Pro plan includes it; the pricing page has the plan details. On Shopify, Mose leaves Shopify's own file alone. The free scan checks whether your /llms.txt exists and opens like one.
The short version
llms.txt is readiness hygiene: a map for the agents and tools that ask for one. Keep it short, true and current, serve it as text, and on Shopify let the platform's own file stand unless you have something better to say. It sits alongside the work that decides whether pages are found and quoted, not in place of it.