Markdown Studio

Free /llms.txt Generator

Enter a domain and this tool reads its homepage, then reads up to 12 pages that homepage links to, and drafts an llms.txt file from what those pages actually say about themselves. Every description is quoted from the page it describes. No account, no limit.

Loading the generator…

What is llms.txt?

llms.txt is a markdown file placed at a site's root that gives a language model a short, structured summary of what the site is and which pages matter. It is a convention, not a standard, and it is read by convention rather than by contract.

The format is a single H1 naming the site, a blockquote summary, then sections of markdown links. It is meant to be read by a machine in one pass, which is why it is plain markdown and not HTML.

What does the generated file contain?

Everything below is read from your own pages at the moment you press Generate. Nothing is written for you.

  • The site title, used as the H1, exactly as your own title tag writes it.
  • Your homepage meta description, used as the blockquote summary.
  • The generation date, as a line of its own directly under that summary, so anyone reading the file can see how old it is before they read anything else.
  • Internal links found in the homepage markup, grouped into sections named after the first segment of each path (Documentation, The check reference, Scoring dimensions, Methodology, Measurement, Tools, Company and Legal), with at most 8 links per section. A section holding more than that states how many of its pages were left out, so the file never reads as though it were complete when it is not.
  • For up to 12 of those links, a description taken from that page's own markup: its meta description, or its H1 where it has none. Nothing is generated, and a subpage never inherits the homepage's description. A link whose page was not read, or whose page carries neither, is listed with no description at all, and the file says how many of those there are.
  • A note on how the file was produced and where the descriptions came from, so a reader can tell what it is.

It does not crawl your site. It reads your homepage and then at most 12 pages your homepage links to, 4 at a time, with a 20-second budget for that whole second phase - so a request cannot take minutes, and a page that is slow or missing is left out rather than guessed at. A page behind a login, or one your homepage does not link to, is not in the file and the file does not pretend otherwise.

How do I deploy it?

Save the output as llms.txt and upload it, then confirm it is reachable.

  1. Copy or download the generated file.
  2. Upload it to your web root so that https://your-domain.com/llms.txt returns it.
  3. Check the response is HTTP 200 with a body, not a redirect to a login page or an HTML error page.
  4. Add a Sitemap: line to robots.txt if it is not already there.

The deployment details, including what a refusal looks like, are in the llms.txt deployment guide.

Is llms.txt a ranking factor?

No. No crawler document we could open commits to reading it — OpenAI’s help centre and Anthropic’s crawler article both specify robots.txt instead. It is worth about 5% of the total score on this site.

The reason it is still worth generating is that it costs almost nothing and the measured gains in the published research come from what a page says and cites, not from adding files:

GEO can boost visibility by up to 40% overall in generative engine responses, and in the paper’s Table 1 the best of the tested methods improve on the no-optimization baseline by 41% and 28%. — Generative Engine Optimization, KDD 2024

That is why the weighting on this site puts citability and evidence at 11% and answer readiness at a further 10%, while the AI context file dimension carries 5%. You can read the whole weighting, and the limits of what it can tell you, on the methodology page.

The shape of the file

Every file this tool produces has the same shape, so you can see what you will get before you run it. The example below is the actual output of a run against this site, with the date it was taken; the links are abridged.

# Your Site Name

> One sentence from your homepage meta description, describing what the site is.

Generated on YYYY-MM-DD. Every link below was read on that date, and a link is only as current as the page it points at: regenerate this file rather than publishing a copy that has aged.

Your Site Name is published at https://your-domain.com. This file lists the pages an AI system should read first.

## Documentation (2 of 4)

- [Guides](https://your-domain.com/docs/): Setup instructions for robots.txt, llms.txt, schema and headings.
- [Methodology](https://your-domain.com/methodology/): The complete scoring method, with every rule.

- 2 of this section's 4 pages are not listed here.

## Company (2)

- [About](https://your-domain.com/about/): What this site is, who maintains it, and what it does not do.
- [Contact](https://your-domain.com/contact/): One address for corrections and data requests.

- 2 listed pages carried neither a meta description nor an H1, so they are linked without a description.

## Notes

- Generated by LLMention from the pages it read on https://your-domain.com. Each description is that page's own meta description, or its own H1 where there is none; no description is written by this tool.
- Review and edit before publishing: the pages are the ones the homepage links to, the descriptions are the pages' own metadata as it stood on the generation date above, and external links are not followed.
- Convention reference: https://llmstxt.org/

Several sections, from a stated rule. A page goes into the section named after the first segment of its own path, and the section order is fixed rather than sorted by size: Documentation, The check reference, Scoring dimensions, Methodology, Measurement, Tools, Company and Legal, then anything else in the order your homepage links to it. Any first segment the tool does not recognise still gets its own section and shows its own name, so what it failed to classify is visible in the file rather than absorbed into a heading nobody can question.

Each section is capped at 8 links and each cap has its own count of what was left out, so a homepage with fifty links produces a short file that is honest about being short. Only 12 pages are read per run, spread one per section per round rather than taken from the top of the page, and a link the tool did not read carries no description rather than an invented one.

The draft is still a starting point. Which pages deserve to be in it, and whether a description your own markup happens to carry is the one you would choose, are your judgement rather than the tool's — and they are the part that decides whether the file is worth serving at all.

How does the scanner score llms.txt?

One dimension out of twelve looks at context files, and it is worth 5 of the 100 points. Four of those five come from the file itself, and one comes from robots.txt being readable at all.

CheckPointsWhat has to be true
llms-txt4The file is more than 20 bytes and contains an H1, a blockquote summary line and at least 3 markdown links. Missing any of those is a partial pass.
ai-context-robots1robots.txt is readable, so some crawler policy is published. An empty body is not a policy.

Note that the check reads the file's contents, not its status code. A 200 response with an empty body scores nothing, and that distinction is deliberate: measuring whether a file is served is not the same as measuring whether it says anything.

Questions about the output

Does llms.txt replace robots.txt?

No. robots.txt states who is allowed to crawl and what they may fetch, and it is enforced. llms.txt is a suggestion about which pages are worth reading, and nothing enforces it.

Where exactly does the file go?

At the site root, so that https://your-domain.com/llms.txt returns it directly with an HTTP 200 and a body.

Will this make ChatGPT recommend my site?

No, and any tool that promises that is overclaiming. Serving the file changes what a model is able to read. It does not change what a model decides to say.

What if my homepage is rendered entirely in JavaScript?

The generator reads server responses and does not execute JavaScript, so a client-rendered homepage gives it little to work with. That is worth knowing on its own, because it is the same response an AI crawler receives.

What the generated file cannot do

Stating the limits is the point of this tool, so here they are for its own output.

  • It cannot know what your site is for. It reads your homepage, so if that page describes a product differently from the way you would describe it, the draft inherits that.
  • It cannot see pages your homepage does not link to, which is usually most of them. That is why every section count is stated in the file: the numbers are what it saw, not what your site holds.
  • It cannot promise that a description it read is still true tomorrow. The generation date is printed at the top for exactly that reason — a file that describes pages as they were on a date is honest, and the same file with no date on it is a claim about the present.
  • It cannot guarantee a citation. Serving a file changes what a model can read, not what it chooses to say.

The scanner this page belongs to is open source, so a disagreement with any of the above can be specific: github.com/Alex13192/geo-scanner. It runs 40 checks across 12 dimensions and publishes every rule it applies, including the ones it deliberately weights low.