What agent-ready documentation actually means
Most teams treat "make us visible to ChatGPT" as a blog problem. It is a docs problem. Answer engines and coding agents read your documentation, your OpenAPI spec, and whatever markdown you bothered to expose. They do not read your brand film.
This is the checklist we send when a founder asks how to get cited, or why their llms.txt is not doing anything.
The failure mode: a link dump
Having an llms.txt is table stakes. A file that is just a list of URLs is not. Agents have limited context and a limited number of tool calls. A pile of links makes them fetch the wrong page, or stop. We have watched this on real docs sites: the file exists, the citations do not.
The same failure shows up as HTML-only docs with no markdown twin, no "Copy for LLM" control, and no way for an agent to pull a single page without a browser.
What agent-ready actually means
Markdown twins for every page. Kernel-style: the HTML is for people, /path.md (or "View as Markdown") is for machines. Put the markdown URL in llms.txt, not only the HTML. If you run a human blog and a crawlable notes repo, keep them labeled as such. Mixing essays with reference is how both audiences bounce.
A "Copy for LLM" control on every docs page. One click, the page as markdown on the clipboard. Agents and humans both use it.
A docs MCP server, not just llms.txt. A well-scoped MCP that an agent installs with the product (the pattern Mastra popularized) beats a root llms.txt for actual tool use. llms.txt remains the index. MCP is how the agent works.
A "For AI agents" block. A copy-paste prompt that does a real job in one shot: migrate this app, swap the base URL, run the quickstart. Two lines. If the agent cannot complete it, your docs are not done.
OpenAPI (or equivalent) that is true. Specs that drift from the product teach models the wrong API. Treat the spec as a test fixture.
Quickstart in more than one shape. Python-only is a tax on everyone else. Add curl and TypeScript at minimum. Put it above the fold, not three clicks into a guide.
The landing page has to match the docs
We keep seeing the same split: docs say self-serve, the marketing site says waitlist. Docs have a base-url swap and a pricing table, the homepage has a screenshot and a "Book a demo." Agents and developers hit both. The contradiction kills trust faster than missing llms.txt.
Put code and numbers on the landing page. If the product is "one command, half the bill," that sentence belongs above the fold, not in a buried changelog.
Separate human writing from machine notes
A technical blog written for people and a crawlable reference written for engines are different jobs. Kernel keeps them apart. We do too: essays live on the insights index, notes are indexed at /llms.txt as markdown twins. Start the notes early. They compound. Do not wait for a perfect CMS.
For how this fits a broader AI DevRel program, see developer relations for AI companies. For the channel mix around it, see alternatives to broadcast social media.
Frequently asked questions
What is agent-ready documentation?
Documentation that a coding agent or answer engine can fetch, parse, and act on without a browser tour. That means markdown twins, an accurate API spec, a useful llms.txt index, and usually a docs MCP, plus a quickstart an agent can complete in one shot.
Is llms.txt enough to get cited by ChatGPT?
No. A link dump does not survive limited context. Use llms.txt as an index to markdown pages, keep the spec accurate, and add Copy for LLM plus a docs MCP if agents are buyers.
What is a docs MCP server?
An MCP server that exposes your documentation as tools an agent can call from the repo or the editor. It is more effective than llms.txt alone for agents that already have the product installed.
How do I make docs visible to AI answer engines?
Expose markdown, structure pages around tasks with the answer in the first paragraph, keep FAQ text identical to the visible FAQ, publish an accurate spec, and do not contradict the docs on the marketing site.
Ready to make the docs the product?
If your landing page and your docs tell two different stories, or your llms.txt is a link dump, start a conversation. This is a docs week, not a six-month program.