llms.txt is a proposed standard, a single Markdown file placed at the root of your domain (/llms.txt), that gives large language models a curated, machine-friendly guide to your most important content and context. It was introduced by Jeremy Howard in September 2024 as a way to help AI systems navigate a site efficiently, much as robots.txt guides crawlers and sitemap.xml lists URLs (llmstxt.org). It is an emerging convention, not an officially adopted requirement, and that distinction matters for how you should think about it.
The problem it tries to solve
LLMs operate within a limited context window, and raw HTML is noisy: navigation, ads, scripts, and boilerplate dilute the actual content. When an AI system wants to understand or reason about your site, it benefits from a concise, structured, plain-text entry point that points it to the pages that matter and explains, in human language, what your site is about. llms.txt is that entry point.
It is deliberately Markdown, not XML or JSON, because Markdown is both human-readable and trivially parseable by models, and it carries lightweight structure (headings, links, descriptions) without ceremony.
What llms.txt is not
A few clarifications that prevent wasted effort:
- It is not
robots.txt.robots.txtcontrols crawler access;llms.txtoffers guidance and context. They serve different purposes and you should maintain both. - It is not a guaranteed input. As of early 2026, major providers have not committed to consuming
llms.txtas a ranking or grounding source. Treat it as low-cost, forward-looking hygiene, not a silver bullet. - It is not a replacement for good on-page content or schema. It points to your content; it doesn't substitute for it.
The format
The spec defines a simple, predictable structure:
# Your Company Name
> A one-line description of what your company or site does.
A short paragraph of additional context: who you serve, what the
site covers, and anything an AI should know to represent you accurately.
## Docs
- [Getting started](https://example.com/docs/start): How to begin.
- [API reference](https://example.com/docs/api): Full endpoint list.
## Guides
- [GEO basics](https://example.com/blog/what-is-geo): Beginner explainer.
## Optional
- [Changelog](https://example.com/changelog): Release history.
The rules are: one H1 with the site/company name, an optional blockquote summary, free-form context paragraphs, then H2 sections containing lists of links with short descriptions. An ## Optional section signals content that can be skipped if context is tight. Each linked page can also have a clean .md companion (e.g. page.html → page.html.md) for a stripped-down version.
How to write a good one
The value is entirely in curation. A dump of every URL is no better than your sitemap. Instead:
- State your entity clearly. The H1 and summary should make it unambiguous who you are. This reinforces the entity-recognition work you do everywhere else in GEO.
- Link only high-value pages. Your best explainers, core service pages, documentation, and authoritative resources, not tag archives or thin pages.
- Write descriptions for the model, not for SEO. One factual clause per link describing what the reader will learn. No keyword stuffing.
- Group logically. Use H2 sections that mirror how someone would reason about your site (Docs, Guides, Products, About).
- Keep it current. Like any map, it's only useful if it reflects the territory. Update it when you publish cornerstone content.
Where it fits in a GEO strategy
llms.txt sits in the same tier as sitemap.xml and robots.txt: cheap, technical, and infrastructural. Because adoption by the big engines is still uncertain, the honest recommendation is to ship one if your site is content- or documentation-heavy, keep it concise and accurate, and not expect it to move citation metrics on its own. The upside is asymmetric: it costs an hour to write and maintain, and if and when more systems consume it, you're already positioned.
The deeper benefit is the discipline it forces: writing an llms.txt makes you articulate, in plain language, what your site is and which pages actually matter. That clarity is the real GEO win, with or without the file.