Yesterday I said off-site content does most of the work in getting you recommended by AI. There’s one big exception, and we’re seeing it a lot more with technical clients, especially ones whose users work with coding agents.
The agent doesn’t search. It just goes to the docs.
What that looks like
Say you’re working in Kubernetes with a coding agent, and you ask it to validate or QA some code. The agent already knows where the Kubernetes documentation lives. So it goes straight there to check its work. It doesn’t go to Google or Bing or Brave. There’s no query fan-out, no search results page, no “best tools” listicle in the way. It’s a shortcut.
Which means for a technical product, the owned vs. earned math flips. Nobody’s weighing your bias against independent sources here. The only question is whether the agent can find the right page and understand it.
This is a lot of traffic
I don’t think most marketers realize how much. Mintlify hosts docs for thousands of developer companies, and in March 2026 they looked at 30 days of traffic. AI coding agents made 45.3% of requests, basically tied with browsers at 45.8%. Claude Code and Cursor were about 95% of that agent traffic. By July, their midyear report had agents at 66%.
And here’s the part that should worry people a bit: you won’t see most of this in Google Analytics. Agents don’t run your tracking script. If your docs are stale, the agent writes broken code with your product, the developer has a bad afternoon, and they just switch to something else. No support ticket. You never hear about it.
Give the agent a map
The simplest fix I know of is an llms.txt at your docs root. Mintlify ran a benchmark with Claude Code and Codex across 20 docs sites, and when agents had an llms.txt to work from, failed lookups basically disappeared.
Beyond that, the stuff I’d look at for a technical client is pretty practical. Serve clean Markdown versions of your pages, since agents parse that a lot better than full HTML. Think about putting an MCP server in front of your docs so agents can query them in the middle of a task. Keep your code samples current, because an old example is the fastest way to lose a developer through an agent. And go look at your server logs, because that’s where this traffic actually shows up.
This is really one piece of a bigger shift I keep thinking about: agents going straight to the source without a prompt or a search at all. That’s tomorrow’s post.
Have a technical product and want to know how agents see your docs? Get in touch.
Leave a Reply