DE
Contact
  • AI
  • Strategy

The Open Knowledge Format, explained: making your knowledge readable for AI

Majd Edriss
Majd Edriss 3 min read

An AI only answers as well as what it knows about you. When someone asks ChatGPT or Gemini about your offering, the model pulls together whatever it finds online – scattered across your website, directories and old posts. Much of it is outdated, contradictory or missing entirely.

This is exactly where a new open format comes in: the Open Knowledge Format (OKF), which Google introduced in the summer of 2026. It sounds like heavy infrastructure, but it is deliberately simple. An honest take on what it is, and what it is not.

What OKF is – in plain terms

OKF is a convention for storing knowledge so that humans and machines can read it equally well. Concretely: a folder of Markdown files, each with a few header lines (title, description, type, tags) and an ordinary text body. The files link to one another and form a kind of knowledge map.

That is the core of it. No database, no server, no special program. Anyone who has seen a Markdown file understands the content right away – and so does an AI agent.

Why this simple approach is clever

The real trick is a decision behind it: OKF is a format, not a product. It belongs to no single vendor, requires no sign-up and no builder tool. That has tangible benefits:

  • No lock-in: your knowledge sits in plain files and can move between systems instead of being stuck in someone else’s tool.
  • Versionable: the files can live in version control like code – with a clear history of who changed what and when.
  • Useful right away: no new platform required. If you have content, you can publish it.

What it explicitly is not

To avoid the wrong impression: OKF is not a replacement for the MCP protocol (which governs how programs talk to a model) – OKF only describes the knowledge itself. It is also not a database and not a silver bullet. As with any tool, the same point applies that I have made about AI in general: it pays off where it solves a concrete problem, not as an end in itself (see When AI is actually worth it).

What this means for small businesses

You do not need large datasets to benefit. The practical step is to make your own website and documentation agent-readable: a compact overview of your key content, cleanly structured, so an AI picks it up correctly instead of guessing. The related, already widespread idea here is llms.txt – a simple content overview for AI clients.

I have put this into practice on this site – as a living example rather than just theory. This website publishes its content automatically in both forms:

Both are regenerated on every release – even this article shows up there shortly after publishing, without me maintaining anything by hand.

My takeaway

OKF is young and still evolving. But the direction is clear: AI systems are increasingly the first place people look, and whoever gives them clean context gets represented more accurately. As so often, starting small and early beats the big project later. I am actively watching the format and building with it.

If you are wondering how your business shows up in AI answers, let’s look at it together.

← Back to blog