What Is llms.txt? A Practical Guide for 2026

llms.txt is a plain-text markdown file, placed at your site''s root, that tells AI language models which of your pages matter and what your site is about. Think of it as a curated briefing for machines: instead of hoping a model lands on the right pages while crawling, you hand it an annotated table of contents at yoursite.com/llms.txt.

The proposal came from Jeremy Howard of Answer.AI in September 2024, and it spread through developer-tool companies first — documentation sites adopted it fastest, since their users were already asking ChatGPT and Claude questions the docs should answer. Since then adoption has crept outward to marketing sites, SaaS platforms, and publishers.

What does an llms.txt file look like?

It''s deliberately simple — markdown, human-readable, no XML ceremony:

# RankCited

> RankCited helps brands get cited by AI engines: visibility tracking plus
> editorial placements on hand-vetted publications.

## Services
- [AI SEO](https://rankcited.com/): the full service, tracking + placements
- [GEO Services](https://rankcited.com/generative-engine-optimization-services): citation building
- [Link Building](https://rankcited.com/link-building-services): vetted publisher marketplace

## Guides
- [GEO vs SEO](https://rankcited.com/geo-vs-seo): how the two disciplines fit together

One H1 with the site name, a blockquote summary, then sections of annotated links. A fuller variant — llms-full.txt — inlines entire page contents so a model can ingest everything in one request. Most sites start with the short version. Ours took about twenty minutes.

Does anything actually read llms.txt?

Here''s the honest part most guides skip. No major AI engine has officially confirmed that it uses llms.txt for retrieval or citations. Google''s John Mueller said as much publicly, and our own crawler-log checks across client sites show visits from some AI bots but no consistent pattern proving the file changes answers.

So why bother? Three reasons that survive the skepticism:

• The cost is trivial. Twenty minutes of writing, no build changes, no risk to anything that already works. As insurance goes, it''s nearly free.
• Adoption usually precedes confirmation. Sitemaps worked this way — publishers adopted them before every engine formalized support. If engines do formalize llms.txt, early files are already in the training and retrieval pipelines.
• The exercise itself is clarifying. Writing one forces you to answer "which ten pages define us?" — and most teams discover their most important pages are buried three clicks deep with vague titles. That finding is worth the twenty minutes even if no model ever reads the file.

llms.txt vs robots.txt vs sitemap.xml

File	Audience	Says
robots.txt	All crawlers	"Here''s what you may and may not fetch"
sitemap.xml	Search engines	"Here''s everything that exists"
llms.txt	AI language models	"Here''s what actually matters, with context"

They complement rather than compete. One practical warning from our audits: teams block AI crawlers in robots.txt (usually via a blanket bot rule someone added years ago) and then wonder why their shiny llms.txt does nothing. GPTBot, ClaudeBot, and PerplexityBot need fetch permission before any file matters.

How to create an llms.txt file

1. Pick the pages that define you. Ten to twenty, not two hundred. Services, flagship guides, pricing, about. If a page wouldn''t help a model answer a question about your category, leave it out.
2. Write one honest line per link. The annotation is the value — "vetted publisher marketplace with transparent pricing" beats "our services page" every time a model summarizes you.
3. Add the blockquote summary. Two sentences, factual, no slogans. Models quote descriptions; write the one you want quoted.
4. Serve it at the root as plain text/markdown: /llms.txt. Static file or route handler, either works.
5. Update it when your site changes shape. A stale llms.txt pointing at dead pages is worse than none — it teaches models exactly the wrong map.

Should you also make llms-full.txt?

Documentation sites: probably, since their whole content base is reference material models should ingest wholesale. Marketing sites: usually not — your pages carry design and persuasion that flatten badly into one giant text file, and the short version''s curation is the real signal. If your docs live on a subdomain, give the docs their own full file and keep the marketing root file short.

The bigger picture

Keep llms.txt in proportion. It''s an on-site signal, and AI engines build brand answers mostly from what other sites say about you — independent coverage outweighs self-description several times over in every citation study we track, including the Princeton GEO research (KDD 2024) that measured a 40% visibility lift for well-sourced third-party content.

So: write the file, spend the twenty minutes, collect the free insurance. Then put your actual effort where the citations come from — earning coverage on the sources engines already trust. That''s the part no text file can do for you.