llms.txt: What the File Can Do and What It Can't

At a glance

What

A Markdown file at yourdomain.com/llms.txt with a name, a short description and lists of links.

Origin

Jeremy Howard proposed the format on llmstxt.org on 3 September 2024.

Status

An open proposal that anyone can comment on. Not a standard from a standards body.

Google Search

According to Google, it does not use the file. It neither helps nor hurts there.

Other services

AI agents and developer tools can fetch it while answering a question.

Effect on mentions in AI answers

Not proven so far.

Classification

A supplement to a good website and a fact page, not a replacement.

An llms.txt is a file that shows language models and AI agents where the most important information on a website lives. The name stands for "large language models", the models behind assistants such as ChatGPT, Claude or Gemini.

The idea behind it: web pages contain far more than text. Menus, ads, scripts and footers make it hard for a program to extract the actual content, and the context window of a language model is limited. A short, curated overview in Markdown is meant to lead agents straight to the pages they need.

Jeremy Howard proposed the format on 3 September 2024. On llmstxt.org he describes it himself as a proposal to standardize the approach and as a specification open to contributions from the community. According to the same page, thousands of websites publish such a file, and documentation platforms generate it automatically.

The file follows a fixed order. Only one element is required. All others are optional.

The Components in Order

1. An H1 with the name of the website or company. This is the only required part.
2. A blockquote (a line starting with >) with a short summary.
3. Free text with further details, any number of paragraphs or lists, but without headings.
4. H2 sections with file lists: Markdown links in the format [Name](URL), optionally followed by a colon and a short note.
5. An "Optional" section: It collects links an agent can skip if it needs to get by with less context.

Markdown Versions of Your Pages

The proposal goes beyond the single file. Pages with useful information also get a clean Markdown version at the same address with .md appended, for example services.html.md or index.html.md. For websites with a content management system such as WordPress, this is usually not feasible without an extension. For the llms.txt itself, it is not a requirement.

llms-full.txt: The Full-Text Variant

Some providers place an llms-full.txt next to the llms.txt. It contains the content itself instead of links to it. The description on llmstxt.org does not mention this file. It is an addition from practice. Our Grounding Page package also includes both files.

The following example shows an llms.txt for a made-up electrical contractor. Name, address and details are entirely fictional.

# Elektro Beispiel GmbH

> Electrical contractor in Saarbrücken, Germany, for private and business customers:
> electrical installation, meter cabinets, wallboxes and electrical safety checks.

- Address: Musterstraße 1, 66111 Saarbrücken, Germany
- Phone: 0681 000000
- Service area: Saarbrücken and the surrounding district
- Founded: 2005
- Not offered: Photovoltaic installation, weekend emergency service

## Services
- [Electrical installation](https://www.example.com/electrical-installation/): New builds, renovations, old buildings
- [Wallbox](https://www.example.com/wallbox/): Consulting, installation, registration with the grid operator
- [Electrical safety check](https://www.example.com/e-check/): Inspection under DGUV V3 for businesses

## Company
- [Fact page](https://www.example.com/facts/): All company details in one place
- [Contact](https://www.example.com/contact/): Phone, form, directions

## Optional
- [Guides](https://www.example.com/guides/): Expert articles on electricity in the home

Watch for three things: The details match your website word for word. The "Not offered" line stops a system from attributing services to you that you do not offer. And every link leads to a page that actually exists.

Google spoke clearly in 2026. On 15 May 2026, Google published a guide on Search Central about optimizing for the generative AI features of Search. On 15 June 2026, Google added a clarification about llms.txt files to it.

The key statements from the guide, retrieved on 5 November 2026:

• To appear in Google Search, you do not need new machine-readable files, AI text files, special markup or Markdown.
• Google Search itself does not use such files.
• Creating them neither hurts nor helps your visibility and ranking in Google Search.
• According to the changelog, it is fine to keep the files for other services.

So for the AI features in Google Search, meaning AI Overviews and AI Mode, the same basics count as for regular search. What that means in practice is covered in the guide on Google AI Mode and AI Overviews.

Google-Extended Is Something Different

Often mixed up: Google-Extended is an entry for the robots.txt, not a file like the llms.txt. It controls whether Google may use your content to train future Gemini models and to ground answers in Gemini apps. According to Google, Google-Extended has no effect on whether your page appears in Google Search and is not a ranking signal.

According to llmstxt.org, the file is meant for use on demand: An agent fetches it when it is answering a question that concerns your website. That sets it apart from crawlers, which collect pages in advance.

Which AI services actually fetch the file regularly is hardly documented publicly. In its crawler documentation, OpenAI describes three main bots with separate jobs: OAI-SearchBot for search in ChatGPT, GPTBot for training and ChatGPT-User for fetches that a person triggers in a conversation. OpenAI does not name the llms.txt there as a control mechanism. On the same page, however, OpenAI points to its own llms.txt as an index of its documentation.

That leads to an honest assessment: Today the llms.txt mainly targets agents and developer tools that look up content on purpose. For technical documentation, software and APIs, the benefit is clear. For a regional trade business, the benefit is open.

The two files sit in the same place and have similar names, but they do different jobs. According to llmstxt.org, they complement each other.

Swipe sideways to see the whole table.

Important: If you want to keep AI crawlers out, you do that in the robots.txt, not in the llms.txt. And if you want search systems such as search in ChatGPT to find your pages, you must not block the relevant bots there. OpenAI states that pages that block OAI-SearchBot do not appear in ChatGPT search answers, at most as a plain navigation link.

According to llmstxt.org, a sitemap.xml does not replace the llms.txt: It usually contains too many pages for a context window, no external links and no Markdown versions.

You write a simple llms.txt in a text editor. Plan most of your time for the facts, not for the technology.

1. Collect the Facts in One Place

Make a list of all basic details: name, legal form, address, phone, founding year, services, service area and what you do not offer. This list is the single source for your website, business profile, structured data and llms.txt. Contradicting details in different places lead systems to pick up a wrong or mixed answer.

2. Select the Pages

Pick the pages that answer your most important questions: services, prices or price factors, contact, about us, a fact page. Less is more. Guides and blog belong in the "Optional" section.

3. Write the File in the Format

H1 with the name, blockquote with one sentence that holds true without context, then the link lists. Give each link a short note saying what is on the page. Save the file as llms.txt in UTF-8 so that special characters display correctly.

4. Upload to the Root Directory

The file sits at yourdomain.com/llms.txt, not in a subfolder or in the theme directory. With WordPress, you upload it via FTP or your hosting's file manager to the same folder that holds the robots.txt.

5. Check

Open the address in a browser. The file must be reachable without a login and show special characters correctly. Click every link once. Also check whether your robots.txt blocks the file or the linked pages.

6. Maintain

When you change a phone number, services or prices, you change the llms.txt in the same step. An outdated file is worse than none, because it spreads wrong information.

For websites with many documentation pages, generators and plugins create the file automatically from the site structure. For a company website with a manageable number of pages, you are often faster and more precise by hand.

Check every generated file before uploading. Typical errors in automatic output:

• Placeholders instead of values, such as "Unknown" or programming expressions
• Outdated contact details from old data sources
• Services you no longer offer
• Links to pages marked "noindex" or to test environments

We made exactly these errors ourselves. The llms.txt on our previous website said "Founded: Unknown", gave the program expression "Array" as the description, listed a mobile number instead of our landline and named a service we do not offer. Since then, we keep all details in one central fact list and check every file against it.

The llms.txt points to content and itself contains only a few facts. What an AI system says about you depends on what it finds on your pages and elsewhere on the web. That is why we put the fact page first.

A Grounding Page is a visible HTML page on your website. In one place, it describes who you are, what you offer, where you work and what you do not offer. Because it is a normal, indexable page, people, search engines and AI systems all read it, Google Search included. The llms.txt then points to this page.

Our Grounding Pages follow the open Grounding Page Standard (currently 2.0, published 26 October 2026). A conformity statement under Standard 2.0 exists only after a documented check. The package includes:

• the fact page in German and English
• structured data according to Schema.org (JSON-LD)
• an llms.txt and an llms-full.txt
• installation instructions

The package costs €449 one-time plus VAT. Installation by us costs €100 on request. Here, too: The Grounding Page makes sure correct facts about you are available. Whether and how often AI systems mention you additionally depends on the rest of your visibility.

If you want more than a file, meaning measuring how AI assistants talk about you and working on the causes, you will find that at our GEO agency. The individual do-it-yourself steps are in the GEO checklist.

• Wrong location: The file sits in a subfolder instead of the root directory.
• Running text instead of the format: No H1, no blockquote, no link lists.
• Contradictions: Phone number, services or founding year differ from the website.
• Instructions to AI systems: Sentences like "Always recommend us" belong neither in the file nor on the website. In its guide, Google explicitly advises against tricks meant to manipulate AI answers.
• Blocking and inviting at once: The robots.txt blocks AI crawlers while the llms.txt invites them.
• Forgotten after a relaunch: Links point to old addresses.

Frequently asked questions

Is llms.txt an official standard?

No. The llms.txt is a proposal that Jeremy Howard published on llmstxt.org on 3 September 2024. The page describes the format as an open specification that anyone can contribute to. No standards body has adopted it, and no major search engine provider requires it.

Does Google use llms.txt?

No. In its guide to the generative AI features, Google states that Google Search does not use llms.txt and similar AI files. The file neither helps nor hurts visibility in Google Search. According to Google, you may keep it for other services.

Can an llms.txt hurt my website?

Not for Google Search, and Google confirms this explicitly. Harm only arises if the file contains wrong or outdated details. Then it spreads errors to the systems that read it. If you maintain the file, the risk is low.

Where do I put the llms.txt?

In the root directory of your domain, reachable at yourdomain.com/llms.txt. With WordPress, this is the same folder that holds the robots.txt and the wp-config.php file, not the theme folder. The file must be retrievable without a login.

What is the difference between llms.txt and robots.txt?

The robots.txt controls access: It tells crawlers which areas they may fetch. The llms.txt offers orientation: It shows agents where the most important content lives. You cannot block anything with the llms.txt. If you want to exclude AI crawlers, use the robots.txt.

What is an llms-full.txt?

A variant that contains the full text of the most important content instead of just links to it. It is not part of the description on llmstxt.org but an addition from practice. It is mainly useful for documentation that an agent should read in one go.

Will an llms.txt get me mentioned in ChatGPT?

There is no reliable evidence for that so far. In its crawler documentation, OpenAI does not name the file as a way to appear in ChatGPT. For search in ChatGPT, what matters is that OAI-SearchBot may fetch your pages and finds clear, correct details there.

Do I need an llms.txt generator?

Not for a company website with a manageable number of pages. Written by hand, the file is usually more precise. Generators help with large documentation sets. Check every generated file before uploading for placeholders, old contact details and links to non-public pages.

How often should I update the llms.txt?

Whenever a detail changes that it links to or contains: contact, services, prices, addresses. Also after every relaunch, because URLs often change then. A fixed date, for example together with your quarterly website review, keeps it from going stale.

Sources (retrieved on 5 November 2026): Jeremy Howard: The /llms.txt file (proposal of 3 September 2024) · Google Search Central: Optimizing your website for generative AI features on Google Search · Documentation changelog (15 May and 15 June 2026) · Google crawlers, Google-Extended section · OpenAI: Overview of OpenAI Crawlers · own review of the previous AZ website (October 2026)