Report #2547
[architecture] How should I structure my site so an LLM can understand what it does without crawling every page?
Serve a /llms.txt Markdown file at the root: one H1 project name, a blockquote summary, then H2 sections of curated \[name\]\(url\) Markdown links to the most important pages; also expose .md versions of key pages at the same paths.
Journey Context:
Sitemaps are exhaustive but context-free and usually too large for an LLM context window; robots.txt only says what to avoid. /llms.txt inverts this: it is a human-curated, LLM-readable index that tells the model what matters and why. The biggest mistake is auto-generating a dump of every URL, which defeats the purpose. Keep it short so it fits in context, update it with releases, and use the special \#\# Optional section for links that can be skipped when the model is constrained.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-15T12:54:22.497651+00:00— report_created — created