Google has released a format for turning a body of knowledge into a folder of linked markdown files. It was designed for internal company data and, by accident, it solves a problem that public websites also have. Right now, the most an AI agent gets from your website is one flat reading of your pagesone at a time. This format creates a graph of how your ideas connect, so I tried it on my own website.
Google’s Open Knowledge format is a directory of linked Markdown files
On June 13, 2026, the Google data team published the open knowledge formator OKF, a way to represent a body of knowledge as a directory of markdown files with a thin layer of YAML frontmatter. Each concept, a table, a metric, a runbook, an API, gets its own markdown document. A short block of YAML contains the searchable fields, type, title, description, resource, tags, and timestamp; the markdown body contains the explanation, and the concepts are linked to each other with ordinary markdown links, which Google says turn the directory into “a relationship graph.” There is no runtime, no SDK, no build step. Google describes a set in three sentences: “just markdown”, “just files”, “just YAML facade”.
The target is the company’s internal knowledge, the context according to Google is “locked behind the surface that created it”, and it is the early version, v0.1, which Google calls “a starting point, not a finished standard”. Nothing in the announcement mentions public websites. This gap is the subject of this piece.
On a Website, a Knowledge Graph Beats Flat Page Copy
The agent-readable version of your website, the one a model or browser actually consumes, is flat. Serve each page as a markdown, the way Cloud Flare done at the edge of the network, is close to AMP for LLMs: a second deleted copy of each page so that a machine can read it. It mirrors what you already have, page by page, and it deletes the same thing with each copy page by page, which is how the pages relate to each other.
A the knowledge graph retains this relationship layer. When your concepts relate to each other, an agent not only learns what each of them is, but it also learns how they relate to each other, which is the bulk of what it really means to understand a website. Two pages can both mention a concept and never tell a machine that one is the frame underneath and the other is the narrower lens next to it. A graph says it clearly, in links that the machine follows. OKF is a standard way to build this graph: markdown, so it’s cheap and structured, so it carries the relationships.
I tried OKF on the No Hacks website
I wrote an OKF bundle for the No Hacks website, a markdown file each for the brand, the host, Machine-centric architecture, the web agentAgent experience optimization, Response engine optimization, llms.txtAnd WebMCP. Each follows Google conventions, YAML fields at the top and a simple markdown body below. The work was mainly deciding which concepts were important and how they connected, not writing the files.
A file, the concept of Machine-First Architecture, looks like this:
---
type: framework
title: Machine-First Architecture
description: A framework for building websites whose full meaning is available to a machine reading them, with the human experience layered on top rather than the other way around.
resource: https://machinefirstarchitecture.com
tags: (Framework, Machine-First Architecture, Agentic Web)
timestamp: 2026-06-13
--- Machine-First Architecture is (Sani)(./sani.md)'s framework for the (agentic web)(./agentic-web.md). The core idea: build the content so a machine reading it gets the complete meaning, the facts, the structure, the relationships, and the human reading gets that same meaning with the design on top. This is why formats that strip a website to plain text, like markdown for agents and (llms.txt)(./llms-txt.md), matter. Its capability side is (WebMCP)(./webmcp.md), and its measurement side is (Agent Experience Optimization)(./agent-experience-optimization.md).
Those bracketed links at the bottom are the chart. An agent following them learns that WebMCP is based on a Machine-First architecture and that llms.txt is the same type of bet, something a flat copy of my pages never says out loud. Through the eight files, it’s the whole structure: the concepts and the relationships between them.
A set like this is a second copy of what the website already says, and a second copy is a second thing to sync. The moment the website changes, the bundle is wrong until you update it too. This tax is not unique to OKF: it’s what each machine-readable parallel layer costs, an llms.txt file, a markdown mirror of your pages, a set like this. The version an agent reads is as accurate as your discipline to keep it up to date.
Google didn’t create OKF for this. Its focus is internal company knowledge, and nothing in its plan points to public websites, so hosting a package for a visiting agent is off-label, and it may stay that way. The reader I created it for, an agent that fetches the packet and follows the graph, might never show up. The reason for doing it has to be valid without that payoff, and it is: writing the bundle forced me to clearly state what No Hacks knows and how its ideas connect, and that exposed gaps I wouldn’t have found writing another page. It’s the same discipline as Machine-First Architecture, put your meaning into a form that a machine can read and you find where you were vague.
Where a Website Knowledge Graph Could Lead
Nothing below is a prediction. It’s a direction, and it depends on agents actually reading the knowledge graphs on websites, none of which are doing so today. The form is always worth the detour.
The identity file could be transformed into a knowledge graph. Today, llms.txt is a single line announcing who you are. A published set is the complete version of this idea, a map of everything your website knows and how the parts connect, so that the thin identity layer and the structured knowledge layer become one.
Agents could query this card instead of deleting your pages. An agent that extracts your packet and follows its links gets a cleaner, relationship-aware read than an agent analyzing your HTML one page at a time, and you have more say in how your own concepts are represented when an AI describes you.
The card could even become the canonical layer. The machine-read version stops being a copy of your website and becomes the source, with human pages just a rendering of it. This is the website designed entirely for machines the Web agent was singled out, accessible via a side door opened by Google for internal data.
Markdown is not new
John Gruber created Markdown in 2004, with Aaron Swartz as a beta tester, and the main design goal was readability: text you can read as is, without rendering, that still converts cleanly to HTML. Two decades later, it runs GitHub, Reddit, much of the documentation you read, and the chat boxes of the AI tools themselves. It won by being readable without being rendered, which is the exact property that makes it easy to read by a machine.
I’ve written most of what I write there for 15 years, since iA Writer became my primary writing app in September 2011, so a week where the agent-readable web converges on markdown is familiar territory for me, not a new trick. The knowledge behind No Hacks (No Hacks OS project) has been working the same way for months: markdown files with structured content, linked to each other, the form that a machine can read and navigate.
Formats for machines keep landing on the same ground, llms.txt, Cloudflare markdown and now OKF. Google itself is not of the same opinion on this subject. Its search side is called llms.txt “purely speculative” for the classification, it is Chrome side added llms.txt check has Lighthouse Agent Readiness Auditand his data team has now released OKF.
If you want to see where your website ranks, it takes thirty seconds. Open your most important page and paste it into a plain text editor, where the links are reduced to single words. Look at what’s left and find anything that indicates how its ideas relate to the rest of your website, not that one page links to another, but the relationship itself. There’s usually nothing, and that absence is what a knowledge graph fills, whether you hit OKF or not.
OKF is this week’s news, and the underlying substrate, plain text that a machine can read, has been there since 2004. What Google added was a standard and a name.
More resources:
This article was originally published on No hacks.
Featured image: Roman Samborskyi/Shutterstock
