# What Is llms.txt and Does It Affect SEO?

> What the llms.txt file is, what llmstxt.org proposes, what Google said in June 2026 about its effect on Search and how to check your store's.

URL: https://clione.ai/glossary/llms-txt/
Lang: en

If someone has offered you "an llms.txt to show up in ChatGPT," it helps to separate two questions that tend to be sold together: **what the file is, which is well defined, and what effect it has on your visibility, which is another matter**. The llms.txt is not [`robots.txt`](/glossary/robots-txt/) under another name or an [XML sitemap](/glossary/xml-sitemap/) for AI. It is a specific proposal, with an owner and a date, and Google has already said what its search engine does with it.

## What llmstxt.org proposes

The proposal was published by Jeremy Howard on September 3, 2024, and a second version came out in August 2026, revised with what was learned in two years of use.[^llmstxt] It starts from a problem anyone who has pasted a website into a chat knows: pages are made for people, they wrap the information in navigation, ads and JavaScript, and turning them back into clean text is hard and imprecise. And a model's [context window](/glossary/context-window/), even though it has grown, **is still too small for most entire sites**.[^llmstxt]

The answer is a file at `/llms.txt` with a fixed Markdown format:[^llmstxt]

- An H1 title with the site's name, which is the only required part.
- A blockquote with a short summary.
- Optional paragraphs with more detail.
- H2 sections with lists of Markdown links, each with an optional note.
- By convention, an "Optional" section for secondary material.

The proposal adds a second piece: that the pages an agent might need have **a clean Markdown version at the same URL**, ending in `.md`.[^llmstxt] Version 2 allows adding `.md` to the full URL or replacing the extension, links those versions with `rel="alternate" type="text/markdown"` and points to the llms.txt with `rel="describedby"`, and clarifies that an llms.txt can sit in a subfolder: it covers the pages under its path, and **the most specific one wins**.[^llmstxt-changes] It is the idea behind [Markdown for agents](/glossary/markdown-for-agents/).

Two clarifications from the proposal itself. The first, what it is for: its author expected it to be useful mainly **at answer time, not for training models**.[^llmstxt] In version 2, an [AI agent](/glossary/ai-agent/) consults the llms.txt to find what it needs and then follows the links it cares about.[^llmstxt-changes] The second, what it is: an informal summary open to community feedback, not a standard approved by any body.[^llmstxt] The proposal cites OpenAI, Anthropic and Google (for the [Gemini](/glossary/gemini/) API) among those who publish an llms.txt for their developer documentation.[^llmstxt] **Publishing your own is not the same as having yours read**, and that is the confusion that moves the most money.

## What effect it has on SEO

Here the answer has a date. On May 15, 2026, Google published its guide to optimizing for the generative AI features of Search.[^google-updates] It says you do not need to create machine-readable files, AI text files, markup or Markdown to appear in Google Search.[^google-ai-guide] On June 15, 2026 it added a note about llms.txt files:[^google-updates] **Google Search does not use them, and creating them neither helps nor hurts your site's visibility or rankings**, because it ignores them. The same note says there is no harm in keeping one for other services or systems.[^google-ai-guide] The guide is about Search with its generative features, so it holds for [AI Overviews](/glossary/ai-overviews/) and [Google AI Mode](/glossary/google-ai-mode/) too.

Chrome treats it as optional. Lighthouse, Chrome's audit tool, has an agentic browsing section with an llms.txt check: **it only flags the page if the server returns an error when the file is requested**. If the file does not exist and returns a 404, the check is marked "not applicable," because offering it is optional for now, and it does not affect the score.[^lighthouse-llms]

For the rest of the assistants, the question is empirical. Each company documents its agents and what they respect, as OpenAI does with [OAI-SearchBot](/glossary/oai-searchbot/), [GPTBot](/glossary/gptbot/) and [ChatGPT-User](/glossary/chatgpt-user/),[^openai-bots] but who actually requests your `/llms.txt`, and how often, only your server's access log tells you.

## In a store

On Shopify you do not start from zero. **Shopify manages a `/llms.txt` by default on stores, which mirrors the content of [`/agents.md`](/glossary/agents-md/)**, the document the platform treats as the main one for agents to discover the store. The lookup order is: an `llms.txt.liquid` template from the theme if it exists, otherwise `agents.md.liquid`, and otherwise the text Shopify generates. The documentation recommends not creating `llms.txt.liquid` unless you need the llms.txt to say something different from `/agents.md`, because then keeping it up to date becomes your job.[^shopify-llms-liquid] Those templates can be customized since May 28, 2026.[^shopify-changelog]

This is what a small photography gear store's llms.txt would look like, in a **fictitious example**:

```markdown
# Example Tripods

> Online store of tripods and ball heads for photography and video, shipping across the United States.

Prices in US dollars. Free returns for 30 days.

## Categories

- [Travel tripods](https://store.example/collections/travel-tripods.md): models under 1.5 kg folded
- [Ball heads](https://store.example/collections/ball-heads.md): maximum load, weight and plate type

## Optional

- [Guide to quick-release plates](https://store.example/pages/quick-release-plates.md)
```

What is useful is in the links: if they point to heavy HTML pages, the file saves little. The proposal wants them pointing to content that is easy for a model to read.[^llmstxt-changes] Clione, for its part, generates an llms.txt and a Markdown document per product, category, collection and page, served on the store's own domain: [how Clione serves files for agents](/features/llms-txt/) tells how.


**The llms.txt neither blocks nor allows anything.** Bots' access rules go in `robots.txt`, which follows the Robots Exclusion Protocol,[^rfc9309] and each company documents which agent respects it and what it uses it for, as OpenAI does with its search agent and its training one.[^openai-bots] If you want your catalog not to be used to train a model, or to appear in an [AI search](/glossary/ai-search/), the place to say so is `robots.txt`. An llms.txt only summarizes and links, and an agent consults it when it needs to.[^llmstxt]


## How to check it

1. **Open `yourdomain.com/llms.txt` in the browser.** If your store is on Shopify and you have never touched it, you will see the text the platform generates from `/agents.md`.[^shopify-llms-liquid]
2. **Read the file the way an agent would**: an H1, a one-line summary and links that work and lead to the pages that actually answer your customers, not to the home page and the legal notice.
3. **Run Lighthouse** in Chrome and look at the agentic browsing section. A server error when requesting the file is the only thing the check flags.[^lighthouse-llms]
4. **Search your server or CDN logs for `/llms.txt`** and note which agents request it. Before believing the name, check the IP: OpenAI, for example, publishes its agents' addresses in JSON files.[^openai-bots]
5. **Do not look for it in [Search Console](/glossary/google-search-console/)**: Google Search ignores the file, so no Google report will show that you have it.[^google-ai-guide]

## Related concept: robots.txt

Both files live at the root and end in `.txt`, and that is where the resemblance ends. `robots.txt` is a protocol published as RFC 9309 by the IETF in 2022, which tells [crawlers](/glossary/web-crawler/) which URLs they may request, and the standard itself warns that **those rules are not a form of access authorization**.[^rfc9309] llms.txt is a reading guide: the proposal distinguishes it from `robots.txt`, which indicates what automated access is considered acceptable, because it is used on demand, when an agent needs information. And it distinguishes it from the sitemap, which does not link model-readable versions and accumulates more than fits in a context.[^llmstxt] One decides who comes in; the other, where to start reading once inside.

The [comparison](/comparisons/llms-txt-sitemap-structured-data/) of llms.txt, sitemap and [structured data](/glossary/structured-data/) puts the three files side by side, and the topic's [reading path](/learn/crawling-and-indexing/) orders the rest of the terms, starting with [indexing](/glossary/indexing/).