What is llms.txt? The complete 2026 guide

What is llms.txt?

llms.txt is a proposed plain-text Markdown file placed at the root of a website (yoursite.com/llms.txt). It is intended to give compatible tools a structured guide to important pages and resources.

Think of it as a curated table of contents for tools that choose to read it. Unlike a sitemap, it is intentionally selective. Publishing the file does not mean a search engine or AI assistant will fetch, trust, prioritize, or cite the listed pages.

The format was proposed by Jeremy Howard, co-founder of Answer.AI, in 2024 in response to the difficulty of fitting large documentation sites into a model's context. It is designed to provide a compact map, but whether that map is consumed depends on the individual tool.

Why does llms.txt exist?

Three considerations motivated the proposal:

1. Context windows are finite. Even modern frontier models with 200K-1M token context windows can't fit a full SaaS documentation site. AI models need help prioritizing what to read.
2. HTML is noisy for LLMs. Navigation menus, ads, footers, and JavaScript pollute the actual content. Markdown is the cleanest format for LLM consumption.
3. Sitemaps weren't designed for AI. They're great for search crawlers but don't signal which pages are most important or what each page contains.

The proposed approach is a single, hand-curated Markdown file that compatible AI agents can read in one request. It is one possible discovery aid, not a replacement for crawlable HTML, internal links, or sitemaps.

The proposed llms.txt format

The proposed format is intentionally simple. Its recommended structure contains, in order:

1. An H1 heading with your site or project name (required, only one allowed)
2. A blockquote summary explaining what your site is about (recommended)
3. Optional intro paragraphs with key context
4. H2 sections grouping related links by category
5. Markdown link lists with descriptions: - [Page title](url): What this page covers

That's it. No XML, no JSON, no rigid schema. Just structured Markdown that an LLM can parse trivially.

A complete llms.txt example

Here's a real-world example for a SaaS product:

# Citegrade

> Citegrade helps SaaS companies and websites get cited by ChatGPT,
> Perplexity, Gemini, and Claude. It scans your pages, identifies why
> AI skips you, and generates one-click fixes.

## Product

- [How Citegrade works](https://citegrade.com/about): Audit pipeline, scoring framework, and what makes content citable
- [Pricing](https://citegrade.com/pricing): Free, Pro ($2.99/mo), and Scale ($9.99/mo) plans
- [Sample report](https://citegrade.com/sample-report): Full audit example with scores, issues, and rewrites

## Features

- [Citation Checker](https://citegrade.com/features/citation-checker): Check if ChatGPT and Perplexity cite your page
- [AI Rewrites](https://citegrade.com/features/ai-rewrites): One-click fixes for citation blockers
- [Readiness Scoring](https://citegrade.com/features/readiness-scoring): 7-dimension content analysis

## Guides

- [How to get cited by ChatGPT](https://citegrade.com/blog/how-to-get-cited-by-chatgpt): 8 changes that increase citation rate
- [Why Google rankings don't mean AI citations](https://citegrade.com/blog/why-your-page-ranks-but-never-gets-cited): Structural reasons LLMs skip top-ranking pages
- [GEO vs AEO vs SEO](https://citegrade.com/blog/geo-vs-aeo-vs-seo): Three optimization disciplines explained

## Optional

- [Changelog](https://citegrade.com/changelog): Recent product updates
- [FAQ](https://citegrade.com/faq): Common questions about pricing, plans, and the audit process

Notice the structure: site name (H1), one-sentence description (blockquote), then categorized link sections. Each link has a brief description that tells the AI what the page contains. The “Optional” section is the convention for content that's helpful but not essential.

llms.txt vs llms-full.txt

The proposal describes two file variants:

• llms.txt — The concise version. 15-30 priority links, designed to fit comfortably in any LLM context window.
• llms-full.txt — The comprehensive version. Includes the full content of every page in Markdown format, not just links. Larger file, but lets the AI ingest your entire site in one request.

If you experiment with the format, the concise file is the safer starting point. A larger llms-full.txt may be relevant to a documentation tool that explicitly supports it, but it increases maintenance and does not create a general search benefit.

How llms.txt differs from robots.txt

They have different roles and very different levels of adoption:

Use robots.txt when you need supported crawl directives. Consider llms.txt only as an optional content map; it cannot enforce access rules or replace security controls.

Who actually reads llms.txt today?

Honest answer: adoption is uneven. As of mid-2026, support must be checked tool by tool:

• Documentation platforms — Mintlify can generate llms.txt files for documentation sites.
• Individual tools and agents — Some may fetch the conventional path or accept it as input, but behavior is not universal. Check the tool's own documentation.
• Google Search — Google says no special AI file is required and that it does not use llms.txt for its AI features.

The honest take: llms.txt is an experiment, not a proven ranking or citation lever. Its clearest benefit today is organizational—it forces a team to identify and describe its most important pages. Any retrieval benefit depends on a tool actually consuming the file.

How to create an llms.txt file in 30 minutes

Here's a fast workflow:

1. List your top 15-20 pages. What URLs best represent your business? Include your homepage, key product/feature pages, pricing, top 3-5 blog posts or guides, documentation, and an “about” page.
2. Write your H1 and summary. One H1 with your brand or product name. One blockquote that explains what you do in 1-2 sentences. Be specific — “a tool for X” not “an AI-powered platform.”
3. Group pages by section. Use H2 headings: Product, Features, Pricing, Guides, etc. Keep groups intuitive.
4. Write descriptive link text. Each link should follow the format: - [Page title](URL): One-sentence description of what the page covers. The description matters more than the title.
5. Save as llms.txt. Place the file at your domain root. For Next.js or static sites, this means putting it in your public/ directory. For WordPress, upload via FTP or a plugin.
6. Verify it's accessible. Visit yoursite.com/llms.txt in your browser. If you see your file rendered as plain text, you're done.

Is llms.txt worth implementing in 2026?

For most websites, it is optional. Here's the tradeoff:

Pros: Relatively quick to create. Forces useful clarity about your most important pages. May help a tool that explicitly supports the proposal.

Cons: Support is inconsistent, maintenance can drift from the real site, and there is no demonstrated Google ranking or AI Overview benefit.

The mistake people make is treating llms.txt as a magic wand. It isn't. If your pages are vague, hedged, and unspecific, no amount of llms.txt curation will get you cited. The file is a map — AI still has to find the destination worth visiting.

That's why llms.txt should be paired with actual citation-readiness work. Audit your pages for the structural and semantic qualities that make AI cite content — or run a sample audit to see what that looks like in practice. Then publish the llms.txt file pointing AI at your strongest pages.