Proposed by Jeremy Howard (Answer.AI) and adopted by companies like Anthropic, llms.txt works as a sitemap for AI. Unlike sitemap.xml (oriented toward search crawlers), llms.txt is a Markdown document readable by LLMs that highlights what's important to know about the site.
Standard structure
# Project Name
> Short description
Optional descriptive text
## Section
- [Link title](URL): description
## Optional
- content that can be ignored if context is limited
Why implement it
- Web-enabled LLMs can retrieve the file directly and index with more precision.
- Explicitly signals which URLs are canonical and priority.
- Reduces hallucination by providing a structured source.
- Emerging standard — early adoption is a competitive advantage in GEO.
Variants
llms-full.txt contains expanded full content. Both can coexist.