llms.txt Generator

What the llms.txt proposal is

llms.txt is a Markdown file placed at the root of a website that tries to tell language models, in a compact form, what the site is and which pages matter. Jeremy Howard published the idea in 2024 on llmstxt.org, and the community keeps refining it.

One distinction matters more than any other: llms.txt is an unofficial, community-developed proposal. It is not an access permission, not an alternative to robots.txt and not a guarantee of visibility in AI systems. No search engine or AI company is obliged to read it. This tool only helps you write a file that follows the format. It is most useful for documentation sites, developer products and teams that publish in-depth guides and want a tidy table of contents.

How to use

1. Enter the site or project name. It becomes the single H1 heading of the file.
2. Write a short summary of one or two sentences; it is added as a blockquote.
3. Optionally add usage notes. Each block separated by an empty line becomes a paragraph; headings are not allowed here.
4. Write the link groups: a ## Heading line starts a group and each line below it follows Title | URL | description. The description is optional.
5. Put pages that can be skipped when context is short into the Optional box; when it has links, a ## Optional section is added at the end.
6. Press Generate, review the warnings, then copy the file or download it as llms.txt and upload it to your site root.

You can also enable an llms-full.txt outline. It lists a heading and a source line for every link and leaves a clear placeholder where you paste the page content yourself.

Link groups and descriptions

The generated file follows the order of the proposal: first # Name, then > summary, then untitled detail paragraphs and finally link lists under H2 headings. Each link is written as - Title: description.

• URLs must be absolute HTTP or HTTPS. javascript:, ftp: and relative paths are rejected and reported with their line number.
• Brackets and backslashes in labels are escaped. Parentheses, spaces and angle brackets in URLs are percent-encoded, so a link never closes early in Markdown.
• Turkish letters in headings and labels are kept exactly as typed, never folded to ASCII. Non-ASCII characters in URL paths use the browser’s standard URL encoding.
• A group without links and an empty Optional box produce no heading; empty groups get a warning.
• Duplicate URLs, plain HTTP links and repeated group headings are flagged as warnings.
• The result also passes a structure check covering the single H1, its position and the list format.

Example and interpretation

Take this input: name Örnek Mobilya, summary Workshop making handmade wooden furniture. and these lines in the link field:

## İçerik rehberi
Care [oil] guide | https://example.com/care_(wood) | Wood care
Products | https://example.com/products

The tool produces:

# Örnek Mobilya

> Workshop making handmade wooden furniture.

## İçerik rehberi

- [Care \[oil\] guide](https://example.com/care_%28wood%29): Wood care
- [Products](https://example.com/products)

The Turkish heading is unchanged. Because the brackets in the label are escaped and the parentheses in the URL are encoded, both links parse correctly. The Optional box was empty, so no ## Optional section appears. Had one line contained javascript:alert(1), no file would be generated and that line would be marked as an error.

How it differs from robots.txt

robots.txt is an established standard that tells crawlers which paths not to crawl, and search engines such as Google document how they interpret it. llms.txt blocks nothing and grants nothing; it is only a summary and a reading list.

If you want to stop an AI crawler, use robots.txt rules; the robots.txt generator offers separate bot options for that. Private content needs passwords or server-side access control. Leaving a page out of llms.txt does not mean it will not be read, and listing it does not guarantee a citation.

Limits and privacy

• Up to 200 links and 100,000 characters of output are supported. Nothing is silently truncated; you get a clear error instead.
• Your pages are not crawled, links are not checked for availability and no .md versions of pages are invented.
• The format is pinned to version v2 of the llmstxt.org proposal as reviewed on 28 September 2026; the tool will be updated if the proposal changes.
• No endorsement by OpenAI, Google or any other provider, no training opt-out and no attribution guarantee is claimed.
• Inputs are processed in your browser and never sent to a server. If you include inputs in a share link, anyone with the link can read them.