A Practical Introduction to Open Knowledge Format (OKF, .v0.2)

SEO

A Practical Introduction to Open Knowledge Format (OKF, .v0.2)

Prosperity Media

24 Aug 2026

Another new acronym has hit the search landscape, raising the same flurry of questions that come along with it: How does it work? How do I implement it? Does this even matter?

Confronted with the same initial questions, I thought I’d break down the technical speak from Google so you can understand what OKF exactly is and how it can help you in your business activities.

Quick note that this article is written against the OKF v0.2 specification, published on 24 July 2026.

Key Takeaways

OKF structures your internal data into simple files that AI agents can natively read.
It relies on basic markdown and frontmatter, requiring no proprietary software.
Version 0.2 introduces strict provenance, trust, and lifecycle rules.
Publishing these bundles will not directly improve your Google search rankings or AI visibility.
In the era of Agentic Search, OKF may play an important role in providing context to agents hitting your site.

What is Google’s Open Knowledge Format (OKF v0.2)

The easiest way to understand OKF is as a standardised way of organising business knowledge so AI agents can find, understand and use it, without requiring any particular software.

A close comparison in my SEO mind is building your own internal version of Google’s Knowledge Graph, mapping specific business entities (concepts) and their relationships. Or, more simply, think of an XML sitemap. Instead of mapping indexable URLs for crawlers, OKF maps your business knowledge for AI agents.

This matters because agents operate best with context. When business knowledge is scattered across documents, tools and fragmented systems, agents are forced to retrieve information and piece it together themselves.

OKF instead creates something closer to a living, structured wiki of your facts, metrics, concepts and processes. This gives agents clearer context, reduces the need to guess, and allows them to maintain cross-references and identify contradictions, while humans guide and validate the knowledge.

Re-Orientating to the SEO/GEO Lens

As search professionals, the first question we ask about any new Google standard is whether it will improve organic traffic or visibility (psst: the answer is no).

While OKF will not directly move the needle on ranking performance, I do think there is genuine value for businesses that choose to package their site as a structured knowledge bundle, especially in the increasingly prominent age of agentic search.

With agent-to-website experiences shifting from basic page scraping to direct context exchange, OKF could serve as the next level of schema markup for LLMs.

Understanding the Bundle Structure

A bundle is physically structured as a directory tree of Markdown files. For example:

sales/
├── index.md
├── datasets/
│   ├── index.md
│   └── orders_db.md
├── tables/
│   ├── index.md
│   ├── orders.md
│   └── customers.md
└── metrics/
    ├── index.md
    └── weekly_active_users.md

The directory structure is independent of the website structure and does not mirror URLs or navigation.

The core methodology relies on the extraction principle. One webpage might contain five distinct concepts, so it becomes five concept files rather than one large file.

Here is a complete and valid minimal concept file:

---
type: Metric
title: Weekly Active Users
description: The count of unique users who logged in over a 7-day period.
tags: [analytics, growth]
generated:
  by: human:jdoe
  at: "2026-08-21T10:00:00Z"
---
# Definition
A user is considered active if they successfully authenticate via the primary login portal.

Every concept file is split cleanly into a YAML frontmatter block and a markdown body (think of the frontmatter as a simple digital instruction sheet for the agent).

Links between concept files are what create the knowledge graph, and a concept can link to another concept or to an external source. The specification recommends structural markdown (headings, lists, tables, fenced code blocks) over freeform prose, because structure aids both human reading and agent retrieval.

What Changed with v0.2?

Version 0.2 introduced four new metadata families in the previously referenced frontmatter block:

v0.2 Family	Detail
Provenance	Records the exact source a concept derives from.
Trust Separation	Separates who generated the content from who verified it.
Lifecycle	Adds status and freshness signals to flag outdated data.
Attested Computations	Provides a sanctioned method for computing specific metrics.

This helps extend the context of agents and build an understanding of trust signals tied to specific signals.

How to Implement OKF

The technical barrier is incredibly low. You only need a directory, a markdown file, and a type field.

Create a directory.
Write one concept file with a type in the frontmatter.
Add a title, description, and tags.
Add generated, verified, and stale_after if an agent will maintain it.
Point an agent at the directory.

If you have some spare time, pop your site into Suganthan’s OKF generator and check out the output.

If you have even MORE spare time, try thinking of a concept (personal or business-related) and working your way back from there. As I mentioned before, the barrier to entry is low, so the only blockers are your own motivation and imagination.

Common Use Cases

There’s a plethora of opportunities that you can tap into with the emergence/development of OKF. The best advice I came across was from Marie Haynes, who recommends uploading the following resources into NotebookLLM and prompting Gemini on how to leverage them within your organisation:

Doing this myself, Gemini made the following case for me:

Deployment Use Case	How to Deploy (Key Steps)	Value
Agent-Ready Website Bundles	1. Group core client expertise into Concept Documents.2. Convert pages to Markdown and cross-link into a relational graph.3. Serve the bundle at /okf/index.md on the client’s domain.4. Point to this directory in the client’s standard llms.txt file.	Maximises brand recall & accuracy. Guarantees that AI search agents ingest highly accurate, structured brand data with zero visual clutter or scraping errors.
Internal Playbooks & Client Audits	1. Document agency SOPs (e.g., Core Update checklists) as OKF Playbook concepts.2. Build a Reference Library of industry standards (Google guidelines, Search Console docs).3. Direct internal agents to load playbooks and run analysis on client data.	Scales agency expertise. Slashes reporting and audit preparation times from days to minutes while ensuring consistent, elite-level execution across all accounts.
Consolidating Data Silos	1. Map out database schemas (like BigQuery), active APIs, and metric definitions (e.g., LTV).2. Standardise metadata with YAML frontmatter.3. Implement OKF v0.2 fields: verified (with actor signatures), status, and stale_after.	Solves fragmented data issues. Acts as a reliable, single source of truth with strict trust tiers, keeping enterprise agents from hallucinating outdated data.

Adoption ready?

So, should you adopt or wait? In short, publishing an OKF bundle will not improve your organic rankings or AI Overview visibility. Google originally built this format for data teams to share internal tables and metrics, rather than for marketers to optimise blogs.

However, if your business is actively deploying internal AI agents, building an OKF bundle provides an immediate, practical way to structure your proprietary data. Additionally, if agentic optimisation is pushed to the forefront of search, OKF could well be positioned to act as the foundational structured data (schema 2.0?) for LLMs.

Why Work With Prosperity Media on AI Search

Navigating new data specifications requires hands-on experience, not just theory. If you are ready to future-proof your digital assets with legitimate generative AI SEO services, we can help. Get in touch with our Sydney team today.