Those of you who follow my work here at the Semantic Blog or have read my book Semantic SEO: Semantic Workflow know that I have a certain pet peeve—a healthy one, to some extent—with the speed at which the digital marketing market invents new acronyms for old problems. In recent months, as soon as large-scale language models gained ground in search engine results, a flood of labels emerged—GEO, AEO, LLMO—as if we had just discovered gunpowder.
The truth, however, is much simpler and, at the same time, much deeper. When I step out of my role as an SEO consultant and put on my Library Science hat, I realize that the central challenge has remained exactly the same for centuries: how to organize, represent, and retrieve information without ambiguity.
It is within this context that it is worth looking closely at a standard proposed by Google, the OKF bundle (Open Knowledge Format). Launched as an open specification in the Google Cloud ecosystem in 2026, OKF is not just another passing trick to manipulate algorithms. In practice, it functions as a portable library, structured in interconnected plain text files in markdown format, created to deliver standardized knowledge to artificial intelligence agents.
Although Google has already made it clear that hosting an /okf/ folder on the server does not function as a direct ranking factor in traditional search, the architectural logic behind this format speaks directly to everything we build in Semantic SEO, which I define as optimization for search engines and information generation.
- What is an OKF bundle in practice?
- How the anatomy of an OKF bundle works
- What is the relationship between OKF, Information Science, and Semantic SEO?
- How to use the OKF bundle in practice within the Semantic Workflow
- Step by step: structuring an OKF bundle in practice
- What do you gain by adopting this semantic discipline?
- A conclusion, for now…
What is an OKF bundle in practice?
Straight to the point, an OKF bundle is a hierarchical package of interconnected Markdown documents that formalizes the concepts, entities, rules, and relationships of a specific knowledge domain.
If we make an immediate bridge to Library Science and Information Science, an OKF bundle is the digital materialization of a lean, machine-readable documentary catalog. Instead of forcing an automated agent to scrape web pages full of menus, banners, scripts, and visual HTML noise, you deliver the semantic core of your project: each concept, product, service, or technical term becomes its own document, with a defined identity and explicit connections to the rest of the collection.
In classification and documentary representation theory, we divide information processing into two complementary fronts:
- Descriptive representation: the attributes that individualize the resource (who created it, when it was updated, what the canonical address is, which category it belongs to).
- Thematic representation: the analysis of the subject (aboutness), that is, what that document is really about, expressed through standardized concepts and semantic relationships.
The OKF bundle addresses exactly these two layers in a single .md file, without requiring complex databases or heavy infrastructure.
How the anatomy of an OKF bundle works
An OKF package follows a directory organization that mirrors the taxonomy of your semantic domain. Within this structure, each file represents a unique unit of knowledge, called a Concept in the specification, and rests on four technical pillars:
1. YAML Header (frontmatter)
At the top of each Markdown file, there is a block delimited by dashes (—) written in YAML. This is where the descriptive representation of the document lives—namely, the entity type, the official title, an objective description, classification tags, the corresponding canonical URL on the web, and governance signals such as update date and authorship. For those who study metadata, it is like having a lightweight cataloging record embedded in the text itself.
2. Structured Markdown Body
Directly below the metadata comes the content itself, written in clean Markdown. Hierarchical subheadings, lists, and tables organize the explanation of the concept. Since large-scale language models process text by breaking it into tokens and calculating statistical and contextual relationships, delivering well-structured prose, without the pollution of visual interface markup, drastically reduces noise during the machine’s reading and comprehension stage.
3. Cross-links (cross-links)
An isolated document is just a loose card in a drawer. What transforms a folder of texts into a navigable knowledge graph are the internal Markdown links between the files in the package. When a product file points to the file of the company that developed it and to the glossary entry that explains the problem it solves, we create implicit semantic triples (subject, predicate, and object) that guide the agent’s reasoning.
4. Manifesto (okf.yaml) and index file (index.md)
No library works without a general catalog to guide the researcher. In the OKF bundle, this role belongs to the okf.yaml manifesto, which declares the identity, version, and entry point of the package, and to index.md, a central map that lists all available concepts. Thanks to this index, an AI agent can consult the catalog first and open only the files necessary to answer a query, saving context window and avoiding hallucinations.
What is the relationship between OKF, Information Science, and Semantic SEO?
There is a common trap in the SEO market: thinking that any new file placed in the domain root (as happened in the race for llms.txt) will generate magic rankings on Google overnight. It won’t. It’s never too much to remind you, before you get too excited: Google has already clarified that the presence of an /okf/ directory is not a direct ranking signal in web search algorithms.
But looking at OKF only through the lens of a “ranking factor” is missing the main point. The OKF architecture mirrors, line by line, the methodology I apply in Semantic SEO projects: organizing knowledge from the inside out, starting from semantic domain modeling before thinking about any published page.
See how the concepts of Library Science, Semantic SEO, and the OKF Bundle overlap:
| Dimension | In Library Science and IS | In Semantic SEO (Web) | In OKF Bundle (AI Agents) |
|---|---|---|---|
| Basic unit | Document or unit of thought (concept). | Web page focused on an entity. | .md file dedicated to a single concept (Concept). |
| Metadata | Descriptive cataloging (MARC, Dublin Core). | Structured data with Schema.org and JSON-LD. | YAML header (type, canonical_url, tags, updated). |
| Terminological control | Controlled vocabulary, thesaurus, and disambiguation. | Unambiguous definitions and entity glossary. | Standardization of preferred terms and synonyms per file. |
| Context and relationship | Hierarchical (genus/species) and associative relationships. | Information architecture and semantic internal linking. | Relative cross-links forming a navigable graph. |
| Discovery | Systematic catalog and subject indexes. | sitemap.xml, taxonomic menus, and breadcrumbs. | index.md, okf.yaml, and reference in the llms.txt file. |
How to use the OKF bundle in practice within the Semantic Workflow
Even without being a direct ranking trigger, adopting the OKF structure brings immediate and lasting gains to any content operation and information architecture. In my daily routine, I see three very clear application fronts:
1. Domain modeling and information architecture before code
In the Semantic Workflow (SWF), we never start a project by opening a keyword volume tool. We start by mapping the organization’s knowledge domain, defining the semantic field, and establishing which entities need to exist on the site.
Building an OKF bundle even before writing the final pages works as a living prototype of your taxonomy. By creating a folder and a .md file for each concept, you immediately see the project’s hierarchy, validate if each page has a unique scope, and eliminate topic cannibalization at the root.
Use the Semantic Suite tools to analyze your knowledge domain and create your taxonomy.
2. Knowledge base for AI content production
One of the biggest mistakes I see companies making today is asking a large-scale language model to write articles or service pages “from scratch,” relying only on the model’s generic pre-training knowledge. The result is almost always superficial text, full of clichés and disconnected from the business reality.
When you feed your editorial workflow (whether in Cursor, Claude, Gemini, or internal agents connected via Git) with your brand’s OKF bundle, the scenario changes completely.
The agent begins to consult your single source of truth, and before writing any line, it reads the unambiguous definitions of your services, respecting the company’s controlled vocabulary and inserting the exact internal links to the correct pages.
Use the Semantic Suite content creation tool to create AI content the right way.
3. Preparation for agentic search and clean consumption by external systems
Companies with extensive technical documentation, complex catalogs, or robust educational bases face a serious problem when AI agents try to read their sites: heavy pages, JavaScript rendering, and fragmented HTML structures make precise fact extraction difficult.
By compiling your public content into an OKF bundle (using tools like the OKF Generator) and pointing to it in the llms.txt file at the domain root, you offer an express, noise-free path for external agents to consume your knowledge base and cite your canonical URLs accurately.
Step by step: structuring an OKF bundle in practice
To move from theory and visualize how this works on the server, let’s set up a complete example for a fictional B2B financial management software company called FinControl.
The goal here is to host the public /okf/ directory at the domain root (fincontrol.com.br/okf/) and connect it to the llms.txt file.
1. Folder structure on the server
The directory tree should reflect the business taxonomy, separating the institutional entity, commercial solutions, and informational concepts from the glossary:
fincontrol.com.br/
├── llms.txt # Points to where the OKF bundle is located
└── okf/ # Root directory of the OKF bundle
├── okf.yaml # Package configuration manifesto
├── index.md # Main index for quick agent reading
├── company/
│ └── about-fincontrol.md # Main brand entity
├── products/
│ ├── financial-erp.md # Pillar page / main service
│ └── pix-reconciliation.md # Specific functionality
└── glossary/
└── cash-flow.md # Informational content (top of funnel)
2. The files that make up the package
A. The manifesto (okf/okf.yaml)
This file establishes the scope, language, version, and entry point of the knowledge base, ensuring that any automated system understands the collection’s boundaries:
name: fincontrol-knowledge-base
version: 1.0.0
description: Official knowledge base on FinControl products, services, and financial concepts.
base_url: https://yoursite.com.br
language: pt-BR
entry_point: index.md
B. The central index (okf/index.md)
Acts as a systematic catalog. It organizes concepts by category and provides a brief scope note for each link, allowing the agent to decide what to read:
—
type: index
title: FinControl Knowledge Index
updated: 2026-11-16
—
# FinControl Knowledge Base
Browse the official concepts of our platform:
## Company
* [About FinControl](company/about-fincontrol.md) – Mission, credentials, and presence in the Brazilian market.
## Products and Solutions
* [Financial ERP](products/financial-erp.md) – Complete management system for small and medium-sized enterprises.
* [Bank Reconciliation via Pix](products/pix-reconciliation.md) – Real-time automated receivables.
## Financial Glossary
* [What is Cash Flow](glossary/cash-flow.md) – Definition, calculation, and management best practices.
C. The concept file (okf/products/pix-reconciliation.md)
Here we see the union between descriptive metadata, clear terminological definition (proximate genus + specific difference, as we teach in Terminology), and cross-links that connect the product to the brand and the theoretical concept:
—
type: product
title: Bank Reconciliation via Pix
canonical_url: https://yoursite.com.br/products/pix-reconciliation/
category: Financial Automation
tags: [pix, bank reconciliation, automation, erp]
updated: 2026-11-16
—
# Bank Reconciliation via Pix
**Bank Reconciliation via Pix** is a module of the [Financial ERP](financial-erp.md) developed by [FinControl](../company/about-fincontrol.md) to automatically reconcile instant charges without manual intervention.
## How it works
* **Automatic identification.** The system reads the unique transaction identifier (EndToEndId) and matches it with open items in accounts receivable.
* **Real-time update.** Bank balance and [Cash Flow](../glossary/cash-flow.md) reports are updated seconds after payment settlement.
* **Native integration.** Connects via API with major Brazilian banks, eliminating the manual exchange of remittance files (CNAB).
## Who it is for
Medium-sized companies with high daily sales volume that need to eliminate operational errors in statement checking.
D. The connection via llms.txt (yoursite.com.br/llms.txt)
Finally, we add the reference to the site root so that agents and crawlers find the official index without having to guess paths:
# FinControl
> Financial management and cash flow automation software for SMEs in Brazil.
## Structured Documentation (OKF)
To access our clean and interconnected knowledge base in Open Knowledge Format, consult the index below:
– [Official OKF Bundle](https://yoursite.com.br/okf/index.md): Contains product specifications, institutional data, and glossary.
What do you gain by adopting this semantic discipline?
When you incorporate OKF bundle logic into your optimization process, three pillars of your digital project are immediately strengthened:
- Precise mapping of canonical URLs: The canonical_url field in the YAML header ensures that any AI agent consuming that concept knows exactly which official web page corresponds to it, increasing the chance of citations with correct and active links.
- Visual cannibalization audit: By transforming your pages into isolated .md files and listing them in a single index, it becomes evident if two articles or service pages are competing for the same concept or explaining the same thing with different names.
- Terminological governance and entity consistency: You define once what your organization is, what the attributes of your products are, and how each term in your domain should be used. Both your human writing team and your AI tools begin to operate on the same controlled vocabulary.
All this brings to your SEO project a richness, depth, and technical rigor worthy of academic research and information organization projects of large companies.
A conclusion, for now…
Tools, protocols, and open specifications come and go. Today we talk about OKF bundles, YAML files, and llms.txt; a few years ago, we talked only about microdata and RDFa; decades ago, librarians were already organizing catalog cards and thesauri with the same logical rigor.
What doesn’t change—and this is what I deeply believe in—is the need to structure knowledge with clarity and without ambiguity. Whether for a traditional search engine to index your pages, for an AI agent to synthesize a reliable answer citing your brand, or for your site’s internal search to semantically retrieve your content, the groundwork remains intellectual: understanding your knowledge domain, defining your entities, and making the relationships between them explicit.
The OKF bundle is an excellent technical bridge to this new phase of information retrieval and generation. It is worth testing, adapting to your project’s reality, and observing how semantic organization transforms the way machines read what you publish.
