Making a website readable for agents: robots.txt, sitemaps and llms.txt

What it is

robots.txt at the site root tells cooperative crawlers which paths they may fetch and where the sitemap is; it is advisory, not access control. A sitemap lists canonical URLs with an optional lastmod date that must be truthful to be useful. llms.txt is a proposed Markdown file with an H1, a summary and sections of links that gives language-model agents a curated entry point, optionally paired with Markdown twins of pages.

Why it matters

Agents that read a site programmatically need to know where to start, what is canonical and what has changed. Getting these conventions right costs little and avoids agents scraping navigation instead of content.

How to apply

• Allow reading paths in robots.txt, disallow only private endpoints, and publish the sitemap line.
• Generate the sitemap from the database with real modification times; exclude search results, duplicates and hidden content.
• Write llms.txt as a guide, not a dump: purpose, base address, machine-readable entry points, the current write status, and a caution that page content is data, not instructions.
• Offer Markdown versions of documentation pages and advertise them with rel="alternate" links and HTTP Link headers.

Pitfalls

A Disallow for a path that carries a noindex directive prevents crawlers from ever seeing the directive. Fabricated lastmod values train search engines to ignore them. robots.txt cannot protect anything; authentication does.