GEO / AI search

llms.txt

Q: What is llms.txt?

llms.txt is a web standard proposed in 2024 by Jeremy Howard for a Markdown file in the root of a website (/llms.txt). It gives large language models a curated, structured overview of the most important content — optimised for LLM readability rather than for human browsers.

Q: Is llms.txt mandatory for GEO?

No — it is optional and not adopted as a binding standard. As of 2026, some providers respect llms.txt hints while others ignore them. The effort is small and the potential benefit positive — which is why it is recommended for small to medium-sized sites and rather secondary for large enterprise sites.

Q: What is the difference between llms.txt and llms-full.txt?

llms.txt provides only structure and links (a short overview). llms-full.txt sets out the most important content in full, as plain Markdown text — LLMs can load the central knowledge base with a single request. llms-full.txt is particularly worthwhile for documentation sites that are consumed by AI coding assistants.

Q: What does an llms.txt look like?

Example structure:# Site name> Short pitch in 1–2 sentences## Important pages- [Feature A](https://example.de/a): What it does- [Feature B](https://example.de/b): What it doesHuman- and machine-readable, usually under 5 KB, giving LLMs a quick overview without HTML parsing.

Q: Does llms.txt replace the sitemap.xml?

No — the two have different jobs. sitemap.xml lists all indexable URLs for classic crawlers (Googlebot, Bingbot). llms.txt curates an LLM-friendly short overview with sentences of context. Anyone who wants to be visible in Google and in AI answers has both files.

Also known as: llms.txt, LLMs Markdown file

llms.txt is a web standard proposed in 2024 by Jeremy Howard for a Markdown file in the root of a website (/llms.txt) that gives large language models a structured, curated overview of the most important content. Unlike sitemap.xml (all URLs), llms.txt provides a contextualised, brief description of the website with links to the central content — optimised for LLM readability. The standard has not yet been bindingly adopted by all major providers, but it is gaining measurable adoption in 2025–2026.

Structure of an llms.txt

An llms.txt is Markdown with a clear structure:

# Website name

> Short pitch — 1–2 sentences on what the site does.

## Important pages
- [Feature A](https://example.de/feature-a): What it does
- [Feature B](https://example.de/feature-b): What it does
- [Glossary](https://example.de/glossar): Collection of terms

## Optional: detail section
[more detailed explanation]

The format is both human- and machine-readable, small (often < 5 KB), and gives an LLM a quick overview of the site structure without having to parse the entire HTML.

llms-full.txt as a supplement

Some websites additionally offer an /llms-full.txt, which sets out not only the structure but also the most important content in full — as plain Markdown text without HTML overhead, optimised for LLM consumption. Advantage: an LLM can load the central knowledge base of a website with a single HTTP request. Disadvantage: the maintenance effort, and the risk of inconsistent content between the HTML and the llms-full.txt.

How Rankmio handles it — and why only with the glossary: At rankmio.com/llms-full.txt you will find every term in this glossary in full text, with definition, context, example and follow-up questions. The feature, comparison and guide pages are deliberately not included. Three reasons:

Only reference material carries without the website. An explanation of a term answers a question even if nobody visits the page. A feature page does not — it is meant to persuade, not to explain. In the full text of an AI file it becomes advertising without an audience.
Structure is lost in plain text. Guide articles depend on images, tables and intermediate steps. As plain text, what remains is a wall of words that is harder to use than the page itself.
Whatever is not generated automatically goes out of date. This is the main reason. The glossary already exists as structured data; the file is rebuilt from it on every request, just like the sitemap.xml. A new or corrected term appears in it without any further action. For the remaining pages, someone would have to maintain the text by hand — and that is exactly where the contradiction between website and file arises that the paragraph above warns about. A file that spreads outdated information does more harm than no file at all.

The price for this is size: the file is around 700 KB, an estimated 200,000 tokens — more than most models will load in passing. That is why the head of the file states its size, and every section begins with the address of the corresponding glossary page. Anyone who only needs one term calls up the page; anyone who wants to read in a reference work takes the file.

Is the effort worth it?

As of 2026, the standard is not bindingly supported by all major providers — how far ChatGPT, Gemini, Claude or Perplexity give priority to llms.txt is hard to measure in any specific case. Recommendation: for small websites (< 100 pages), an llms.txt with the 10–20 most important links and short descriptions is inexpensive effort with a potential GEO bonus. An llms-full.txt is worthwhile above all for documentation websites that are explicitly intended to be consumed by AI-supported tools (developer docs, API references).

Example from practice

Example: Rankmio itself. At rankmio.com/llms.txt there is a file of around 12 KB — short enough for a language model to read it in full while answering a question. It begins with one sentence on what Rankmio is, then names the data sources, the pricing model with real figures and a paragraph on what Rankmio is not — the paragraph that helps most, because otherwise a model assigns an unfamiliar tool to the nearest larger category. After that come 67 references in seven groups, each with its own descriptive sentence written by hand rather than pulled from the meta tag.

The file is not generated once but on every request: new glossary terms are added automatically, and the German edition at rankmio.de comes from the same source. The date at the foot is taken from the modification date of the content, not from the clock — a date that changes daily without any real change would be a freshness signal that is not true.

Frequently asked questions

What is llms.txt?

llms.txt is a web standard proposed in 2024 by Jeremy Howard for a Markdown file in the root of a website (/llms.txt). It gives large language models a curated, structured overview of the most important content — optimised for LLM readability rather than for human browsers.

Is llms.txt mandatory for GEO?

No — it is optional and not adopted as a binding standard. As of 2026, some providers respect llms.txt hints while others ignore them. The effort is small and the potential benefit positive — which is why it is recommended for small to medium-sized sites and rather secondary for large enterprise sites.

What is the difference between llms.txt and llms-full.txt?

llms.txt provides only structure and links (a short overview). llms-full.txt sets out the most important content in full, as plain Markdown text — LLMs can load the central knowledge base with a single request. llms-full.txt is particularly worthwhile for documentation sites that are consumed by AI coding assistants.

What does an llms.txt look like?

Example structure:

# Site name
> Short pitch in 1–2 sentences
## Important pages
- [Feature A](https://example.de/a): What it does
- [Feature B](https://example.de/b): What it does

Human- and machine-readable, usually under 5 KB, giving LLMs a quick overview without HTML parsing.

Does llms.txt replace the sitemap.xml?

No — the two have different jobs. sitemap.xml lists all indexable URLs for classic crawlers (Googlebot, Bingbot). llms.txt curates an LLM-friendly short overview with sentences of context. Anyone who wants to be visible in Google and in AI answers has both files.

Used in Rankmio for

GEO audit: checks page by page how citable it is for a language model

Go to the feature →

Last updated: 2026-08-20 · Browse all glossary entries