What Is the Open Knowledge Format (OKF)? Google's New Standard and Where It Sits in the GEO Picture

The short version: OKF is a packaging format for knowledge, not a new SEO tag

People have started forwarding us Google’s Open Knowledge Format (OKF) announcement, asking “should we rush to add OKF to our site?” Let’s be blunt up front:

OKF is a file format for packaging knowledge so that both humans and AI agents can read it. What it set out to solve is the problem of organizational knowledge scattered across incompatible systems, where every AI agent builder has to reassemble context from scratch — it is not a tag you paste on a marketing site to get ChatGPT mentioning you more this week.

In Google’s own words: a bundle “will not move your rankings or your AI visibility this week.” Its real value is directional: the day an agent comes looking, your knowledge is already organized and “effortless to read.” So this post isn’t telling you to rebuild your site today — it’s helping you understand where OKF sits in the GEO picture and how it relates to what you’re already doing.

What OKF actually is

In June 2026, Google Cloud’s Data Analytics team (Sam McVeety and Amir Hormati) released OKF v0.1. It formalizes the emerging “LLM-wiki” pattern articulated by Andrej Karpathy: rather than having an AI re-search the same documents every time, you give it a shared Markdown knowledge library that grows more useful over time, with the agent handling the tedious cross-reference bookkeeping.

Technically it’s deliberately thin:

• Just Markdown — opens in any editor, renders on GitHub, is indexable by search tools
• Just files — shippable as a tarball, hostable in git, mountable on a filesystem
• Just YAML frontmatter — a small set of queryable structured fields

One concept = one Markdown file, and the file path is that concept’s identity. Each file declares a few YAML fields at the top (only type is required):

---
type: BigQuery Table
title: Orders
description: One row per completed customer order.
resource: https://console.cloud.google.com/...
tags: [sales, revenue]
timestamp: 2026-05-28T14:30:00Z
---

# Schema
| Column | Type | Description |
|--------|------|-------------|
| `order_id`    | STRING | Globally unique order identifier |
| `customer_id` | STRING | FK to [customers](/tables/customers.md) |

Notice the last line — a Markdown link pointing to another concept file. That’s the key thing OKF adds over a pile of loose documents: the relationships between pieces of knowledge are preserved, not just the page-by-page content but how the pages connect.

The problem it solves (the most common misunderstanding)

Glance at the fields and you’ll see it: BigQuery Table, table schemas, console links — OKF’s first home is internal enterprise data and knowledge, giving AI agents inside an organization consistent, trustworthy context instead of each one reassembling its own.

So if your question is “should my marketing site adopt OKF,” the honest answer is: as designed in v0.1, it isn’t aimed at your public-facing website. No mainstream AI engine has announced that it will crawl public OKF bundles to decide whether to cite you. Saying this plainly matters, because the distinction directly affects whether you should invest time now.

So does it relate to GEO and being cited by AI?

Yes — but directionally, not as this quarter’s tactic.

Zoom out and the “machine-readable web” has been stacking up layer by layer, each answering a different question:

What’s genuinely worth noting is that OKF shares the same instinct we keep emphasizing: AI doesn’t just look at how polished a single page is — it looks at whether it can understand you as a consistent, internally connected knowledge entity. OKF bakes the page-to-page relationships into the format itself, which is the same road as our work on the brand knowledge system and entity consistency.

But hold onto this: OKF is v0.1, adoption is still unknown, and the public-web side is even less formed. It’s a direction worth tracking, not a “do it now or you lose” to-do. Anyone using it to scare you into rebuilding your site is dressing up something rare as if it were the foundation of your site’s health — the order is exactly backwards.

Do we already have anything related?

Yes, quite a bit. The layer OKF describes is an extension of GEO fundamentals you’re already doing or should be:

• llms.txt — a plain-text source of facts and a signpost for AI crawlers; see SSR vs SPA: the rendering problem behind AI citation
• Structured data (JSON-LD) — author-declared facts the engines trust; see How structured data affects LLM citation
• Agent readiness — where llms.txt, MCP servers and other emerging standards sit in the 12 GEO dimensions; see How GEO dimensions interact
• Brand knowledge system — turning scattered brand facts into a consistent entity AI can read; see The GEO brand knowledge system

OKF isn’t a new job category out of nowhere — it pushes this “organize it for machines” work one notch further toward being more formal and more portable.

So what should you actually do now?

Don’t: tear down and rebuild your site to produce OKF bundles for a v0.1 standard before any mainstream engine has committed to consuming it. That’s chasing hype, not building site health.

Do: exactly the fundamentals that pay off now and also make you “OKF-ready” later:

1. Make sure AI crawlers can actually fetch your content (pure client-rendered sites often hand them an empty div)
2. State who your company is, what it does, and your track record consistently across your site and beyond
3. Keep internal links and topical relationships clear, so “how pages connect” already holds
4. Get structured data right and consistent with the content (mismatched markup costs you)

Get these solid and the day a mainstream engine starts consuming such knowledge bundles, you’re already organized and waiting rather than scrambling. And these four aren’t a one-off project — the AI engines, your content, and your competitors all keep moving, so it’s site-health work you have to keep watching and maintaining. That’s exactly why we run it as continuous monitoring and managed service: doing it yourself along these lines is completely fine, it’s just that watching it by hand every week will burn you out.

OKF is worth knowing exists and knowing which slot it fills. But don’t mistake “knowing a new standard exists” for “having to act on it now” — keep the foundation in shape, and when the new standard arrives you’ll already be standing in the right place.