llms.txt

llms.txt is a plain-text file, placed at the root of a website (/llms.txt), that gives AI systems a curated map of a site's most important pages — written for a language model to read, not for a human visitor to stumble on. Think of it as a sitemap.xml aimed at Claude and Perplexity instead of Googlebot.

The standard was originally proposed by Jeremy Howard (fast.ai) and is supported by Anthropic's Claude. The file is formatted in Markdown and provides a hierarchical list of the website's most important pages with brief descriptions:

# Business Name

> Brief description of the organisation and what the site covers.

## Core pages
- [Page title](/url): Brief description of what this page covers

## Resources
- [Resource title](/url): Description

## Optional: additional information for LLMs

Why it exists

Language models process text in tokens, and a full website — with navigation menus, cookie banners, footer boilerplate, and marketing copy — burns a lot of those tokens on content that has nothing to do with the actual question a user asked. llms.txt strips that away. It tells a model: here is what this business actually is, and here are the dozen pages that matter, skip the rest. For a model deciding whether to cite a source, that's a much cheaper way to confirm relevance than crawling the whole site and figuring it out.

Adoption status

• Perplexity: confirmed to use llms.txt
• Anthropic (Claude): developed the standard, uses it
• OpenAI (ChatGPT): partial/unconfirmed
• Google: not confirmed to use llms.txt for AI Overview selection

This list moves. It's worth checking again every few months rather than treating any of the above as permanent — AI crawler behaviour is still one of the least stable parts of the search stack.

Value for AEO

llms.txt is a low-cost, forward-compatible implementation. It helps AI systems that honour it navigate to the most relevant pages efficiently, and has no negative effects on systems that don't use it — an unusual property for an SEO tactic, since most either help or do nothing, and this one genuinely can't hurt. It's also fast to build: unlike structured data →, which needs to be applied page by page, a single llms.txt file covers a whole site in one document.

What llms.txt is not

It's not a ranking factor, and there's no confirmed evidence it directly influences whether Google includes a page in an AI Overview. It's not a replacement for LocalBusiness schema →, which declares structured facts about a specific business — llms.txt is closer to a table of contents. And it's not something a model is obligated to respect; unlike robots.txt, which crawlers generally honour by convention and by law in some jurisdictions, llms.txt compliance is voluntary and currently limited to a handful of providers.

Common mistakes

The most common mistake is treating llms.txt as a dumping ground — listing every page on the site rather than curating the dozen or so that actually represent the business well. A file with 200 entries defeats the purpose; the value is in the editorial judgment of what matters most. The second mistake is writing the descriptions like SEO meta descriptions stuffed with keywords, when the audience here is a model trying to understand what a page contains, not a search engine trying to rank a snippet. Plain, accurate one-line descriptions work better than persuasive ones.

India context: Very few Indian business websites have implemented llms.txt as of 2026. Early implementation is a differentiator for businesses targeting Perplexity and Claude citations — a Jaipur-based hospital chain or a Pune SaaS company that ships one now is doing something almost none of its competitors have done yet, for the cost of writing one file.

Related terms: AI Crawler → · GEO → · LLM Citation → · LocalBusiness Schema → · Structured Data → · llms.txt Setup Service →

Example: Angryturtle's own llms.txt lists its most important AEO service pages, glossary, the DCG framework explanation, and industry pages — guiding AI crawlers to the highest-value content rather than letting them discover pages through standard crawling and guess at what matters most.