Skip to main content
Your content has two audiences: humans and AI. Not long ago, someone who needed help with your product would start with your docs. If they couldn’t find an answer, they might open a support ticket or try another product. That pattern has changed. People now ask tools such as ChatGPT, Claude, Perplexity, and your Mintlify assistant for answers grounded in your content. Developers also use coding agents such as Claude Code and Cursor to understand and build with your product. Your content must still work well for humans. But you also need to consider the AI that mediates between your content and your users. The good news is that agent-friendly content is usually better for people too. Clear pages, explicit context, consistent terms, and a tight focus help every reader. This course also covers the few considerations that are specific to agents.

​
What this course covers

  • How agents read your content — What happens when an agent processes your content, and why some content works better as context for agents.
  • Write content agents can use — The writing and content design choices that make your content reliable context for agents.
  • Control what agents see — Which settings and files to configure, including llms.txt, skill.md, CLAUDE.md, and AGENTS.md.
  • Test whether agents can use your content — How to evaluate answers, citations, completeness, and restraint with a repeatable question set.
  • Protect private and sensitive information — How to separate discovery from access control and keep secrets out of agent context.
  • Keep content agent-friendly over time — A maintenance approach for a fast-changing AI landscape.

​
Who this is for

Developers, technical writers, and product teams who maintain content. Some lessons focus on Mintlify, but the principles apply to any platform. You don’t need a background in AI or machine learning, just an interest in making your content work well for anyone (or anything) that reads it.

​
Lessons in this course

  1. How agents read your content — What happens when an agent processes your content, and what determines whether it succeeds
  2. Write content agents can use — Techniques for writing content that serves as reliable context for agents.
  3. Control what agents see — Configure what agents have access to, and give them the context to produce accurate answers.
  4. Test whether agents can use your content — Build a repeatable evaluation set for accuracy, grounding, completeness, and restraint.
  5. Protect private and sensitive information — Separate discovery from access control and test what each audience can retrieve.
  6. Keep content agent-friendly over time — How to maintain agent-friendly content as your product grows and best practices evolve.

​
Agent-friendly content checklist

Use this as a quick reference.

​
Content

  • Every page makes sense without reading adjacent pages
  • Page titles are specific enough to match user queries (“Configure rate limits for API requests” instead of “Advanced configuration”)
  • Page descriptions answer “why would someone read this?” with specific topics. No vague summaries
  • Headings make sense without surrounding context and tell a narrative from just skimming them
  • Each page covers one topic or task, focused on a single user goal
  • Terminology is consistent throughout with one name per concept

​
Examples and references

  • Code examples are complete and runnable
  • Placeholder values are clear and explain what to substitute
  • Cross-references name what they link to. No “click here,” “the above,” or similar

​
Configuration

  • Your site has an llms.txt file
  • llms.txt has a blockquote description that summarizes the content of your site
  • Each entry in llms.txt has a description and a link
  • Important pages appear first in llms.txt
  • Deprecated, internal, and changelog pages are excluded from llms.txt
  • AGENTS.md or CLAUDE.md defines audience, terminology, and content type rules for coding agents
  • Private content is protected by authentication and authorization

​
Evaluation

  • A fixed set of user questions covers lookup, tasks, troubleshooting, ambiguity, and unsupported requests
  • Expected facts and supporting sources are recorded for each question
  • Evaluations check correctness, completeness, grounding, restraint, and usefulness
  • The evaluation set runs after major content, terminology, configuration, or access changes

​
Ongoing maintenance

  • Review user conversations with agents regularly for failures and wrong answers
  • Update content alongside product changes
  • Review llms.txt when major sections are added or removed, unless your llms.txt automatically updates