Modern AI assistants — such as ChatGPT, Claude, Gemini, etc. — increasingly retrieve and summarize content from the web in real time. To serve these tools accurately, your documentation portal needs to expose content in a way that AI agents can efficiently discover and read.
What makes a documentation portal AI-friendly:
- Using llms.txt — a curated, human-readable index file that tells AI agents which pages in your portal are the most authoritative and relevant. It follows the llms.txt convention proposed by the AI research community and is analogous to robots.txt and sitemap.xml for traditional search engines. Learn how you can configure llms.txt for your ClickHelp portal.
- Machine-readable formats — providing documentation in Markdown or via API (JSON) instead of PDFs or complex HTML markup, which bots may not always be able to parse. With ClickHelp, every topic in your portal can be retrieved as clean Markdown. This allows AI agents and LLM-powered tools to read individual topics without parsing HTML.
- MCP Server — a Model Context Protocol (MCP) server that allows AI coding assistants and agents to query your documentation programmatically.
- Semantic markup — a strict hierarchy of headings (H1 → H2 → H3) that helps models correctly "chunk" the text for analysis.
- Optimal topic length — keep individual topics neither too short nor too long. Very short topics lack the context AI needs to answer questions accurately; very long topics may exceed the context window of LLM-powered tools or cause relevant content to be diluted. This aligns with the principles of topic-based authoring: each topic should be self-contained and focused on a single concept—which turns out to be equally good advice for both human readers and AI agents.
- No "hidden" context — AI cannot guess. Terms should be clearly defined in the glossary, and code examples should contain all necessary imports to run immediately.
These tips work together to ensure your documentation surfaces correctly in AI-generated answers, developer tools, and agentic workflows.