llms.txt is a 2024 proposal by Jeremy Howard (Answer.ai) for a standard file at /llms.txt declaring to LLMs which site content is available, how it is structured, and which canonical version to prefer. An attempt at a “robots.txt” adapted for the LLM ecosystem. We implement it as part of our agentic search work.
What it declares
Markdown format with clear sections:
- Site name and description.
- List of URLs with purpose and brief summary.
- Hierarchical structure (categories, taxonomies).
- Reference to LLM-optimised versions if they exist (e.g.
/api/llm/).
Adoption status (April 2026)
- Anthropic (Claude): declares support, respects the file.
- OpenAI (ChatGPT, GPT): no explicit official support; anecdotal evidence it reads the file.
- Google (Gemini, AI Overviews): no official statement. No evidence of direct use.
- Perplexity: explicit support.
How we implement it at SEOCOM
For clients with active GEO presence:
- List critical URLs (pillars, main products/services, structured resources).
- 1-3 line summary per URL explaining its value.
- Link to plain MD or TXT versions where available (easier for an LLM to parse).
- Keep the file updated when new important resources ship.
Not yet an official standard. But implementation cost is low, downside non-existent, and potential upside significant if LLMs formalise support.