Mintlify Is Hot With Humans. AI Agents Barely Pick It.

Mintlify is everywhere in dev-tool circles right now. Beautiful docs, AI-era pedigree, a customer list full of the companies you'd expect. If you asked a room of developers which docs platform is hot, plenty would name it.

So we asked the other audience that matters now: AI agents. We ran unbranded tasks — "set up docs for my project," "generate an API reference from my OpenAPI spec," "what documentation platform should I use" — across Claude, GPT, and Gemini, and scored which tool actually got picked. The result is a clean illustration of something every new dev tool should worry about.

The buzzy names lose to the old frameworks

The whole category is fragmented — even the leader sits around a 10% Pick Rate, and most tools are in the low single digits. But the ordering is the story. The tool agents reach for most is Docusaurus (~10%), the oldest and most ubiquitous open-source docs framework, followed by the OpenAPI reference generator Redocly (~8%) and the markdown frameworks Scalar and Nextra (~6%). The hosted, well-marketed names sit behind them — Mintlify around 3%, with ReadMe in the low single digits and GitBook barely registering.

One number cuts deeper than the Pick Rate: Shortlist Rate, how often a tool is named at all. Docusaurus is named in well over half of trials — agents know it cold. The hosted names are named a fraction as often. Agents aren't weighing Mintlify and passing; a lot of the time it doesn't come to mind at all.

See the current ranking on the documentation platforms leaderboard — it updates as we re-run.

This is the Awareness gap, not a quality verdict

Mintlify isn't losing because it's worse. It's losing because of the first stage of the Agent Funnel: Awareness — corpus presence. An agent recommends what it has seen most, and what it has seen is the public training data: years of GitHub stars, Stack Overflow answers, tutorials, and — the big one — other projects' docs built on these frameworks. Docusaurus has been in a million repos since long before the current models were trained. A newer hosted platform, however good, simply hasn't accumulated that footprint yet.

That's the trap, and it's not unique to docs. Human buzz and agent mindshare are different markets. You can win every conference hallway and still be invisible at the moment an agent writes the scaffold. The mailing-list darling loses to the boring, ubiquitous incumbent because the incumbent is what the model absorbed.

Why it compounds

Docs are a special case of the corpus problem because they feed it. Every project that builds on a framework publishes more docs, more config, more examples — which become more training data for that framework, which makes agents recommend it more, which wins it the next project. The open-source frameworks are in a flywheel. The hosted platforms have to break in from outside it.

How a new tool closes the gap

The gap is real but not permanent. Closing it is the work of Agent SEO:

• Get into the corpus. Be written about where models train — clean, crawlable docs, real GitHub presence, examples other people can copy. Slow, but it's the foundation.
• Be machine-readable today. An llms.txt, clean HTML, and markdown endpoints decide whether an agent can even read you in real time. (Pickrate's own docs are agent-readable — append .md to any page.)
• Win the unbranded prompt. Find the tasks you lose — "generate an API reference," "set up docs" — and the rival taking them, and earn the recommendation there.

The docs category is the cleanest example we've measured of a market where the hyped product and the agent-recommended product are different products. If you're building something new, that should worry you — and if you're measuring it, it's exactly where the opportunity is. Check your own Pick Rate and find your Awareness gap.