Making your docs readable by AI

Customers ask AI assistants about your product before they visit your site. Your help center publishes machine-readable knowledge so those assistants quote you accurately instead of guessing.

None of this needs configuring. Publishing your help center is enough.

llms.txt

https://your-slug.helpshelf.com/llms.txt

A compact map of your knowledge — roughly 10KB — listing what you've documented and pointing to the full bundle and your MCP endpoint. It's small enough to drop into a prompt whole.

Under AI Agent Access it advertises the other two surfaces, so an assistant that finds this file can discover everything else.

llms-full.txt

https://your-slug.helpshelf.com/llms-full.txt

Your complete verified context bundle, capped at around 300KB.

It includes curated and standard content only. AI-generated drafts are excluded — external assistants never quote back unreviewed AI text as though it were your documentation. See Trust tiers.

Crawler policy

Your help center serves a robots.txt that explicitly welcomes AI crawlers — GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot and others — alongside conventional search engines. A sitemap is published too.

If you put your own CDN or proxy in front of a HelpShelf URL, check its bot settings. Managed "block AI crawlers" rules are common, on by default at some providers, and will override the policy your help center serves. The symptom is an assistant reporting it can't access your documentation even though the URL loads fine in a browser.

npx helpshelf score catches this, since it checks from the outside. See AI legibility score.

Ask with my AI

Your help center and widget both offer an Ask with my AI button. It copies a ready-made prompt referencing your llms.txt, or deep-links into ChatGPT, Claude, or Perplexity with it.

This turns a visitor who was going to ask an AI anyway into one asking with your real documentation attached.

For live queries, use MCP

llms.txt is a snapshot an assistant reads. For agents that need to search your knowledge on demand, the MCP server is the better surface.