What is llms.txt? (The plain-language guide)
llms.txt is a small markdown file you place at the root of
your website. Its job is simple: hand large language models a clean,
curated map of what your site is and which pages matter — instead of
making them guess from messy HTML, navigation menus, cookie banners and
ads. Think of it as a README for AI.
Why it matters for AI search (GEO) in 2026
Search is splitting in two. Alongside the classic blue links,
people now ask ChatGPT, Claude, Perplexity, and Google’s AI
Overviews — and those tools summarize the web instead of
just listing it. Generative Engine Optimization (GEO)
is the practice of making sure that when an AI answers a question in
your space, it understands you correctly and cites you. A clear
llms.txt is one of the cheapest, most direct GEO moves
available: you’re literally writing the description the model reads.
How AI crawlers use it
When an AI system ingests or browses your site, an
llms.txt at the root gives it a concise, token-efficient
summary plus a hand-picked list of your most important URLs and what
each one is. That reduces the chance of the model hallucinating about
you, missing your key pages, or describing you with stale or
third-party info. It complements — it doesn’t replace — your real
pages: the model can still follow the links to read the full content.
Is this an official standard?
llms.txt is an emerging, community-driven
convention (proposed in late 2024), not yet a formal
requirement enforced by any single AI provider. Adoption is growing
fast across docs sites and tools. Treat it like an early best
practice: low effort, no downside, and increasingly expected. The
format below follows the de-facto spec.
llms.txt vs robots.txt vs sitemap.xml
Three different files, three different jobs. You’ll often want all three.
The short version: robots.txt says who’s
allowed in. sitemap.xml lists every door.
llms.txt tells the AI what the building is and which rooms
are worth visiting.
Anatomy of a good llms.txt
- H1 brand name — one
# Your Brand at the very top.
- Blockquote summary — one
> line the AI can quote verbatim.
- Optional context — a short paragraph or two of plain prose.
- H2 sections —
## Docs, ## Products, ## About, each a bullet list of - [Title](URL): description.
- An
## Optional section — by convention, lower-priority links go here.
The generator above produces exactly this structure. Edit the seeded example and watch the output update live.