Check your company’s visibility in AI
Free, with no credit card required.
A customer asks about the terms of a service. The website offers a general description, a guide contains an older version of the offer, and the instructions mention an exception they had not noticed. Each resource may look correct, but together they fail to give a clear answer. An AI system using this content also has to resolve those differences.
That is why preparing knowledge for AI starts with agreeing on what the information means. What exactly are we describing? When does a condition apply? Where is its source? Open Knowledge Format helps record this knowledge in a structured, portable form.
What is Open Knowledge Format?
Open Knowledge Format, or OKF, is an open format for recording knowledge for people and AI systems. Google Cloud introduced it on 12 June 2026. It starts with a simple idea: knowledge can be stored in text files that different tools read according to shared rules.
The foundation is Markdown documents: text with simple notation for headings, lists and links. A YAML block at the beginning contains metadata - short fields describing the content. A collection of these files can represent metric definitions, data descriptions, procedures or service rules.
The format does not require a particular cloud or AI model. Its purpose is to make knowledge exchange between tools easier. Recording information in OKF does not, on its own, mean that an external system has retrieved or used it.
Source: Google Cloud - introduction to Open Knowledge Format
Google’s article describes version 0.1. What changed in version 0.2?
Google’s article is a useful introduction, but it covers the initial version 0.1. The specification checked on 13 October 2026 already lists version 0.2. The project also has its own repository. The old copy in knowledge-catalog/okf is no longer maintained.
Version 0.2 expands the description of knowledge provenance, verification and currency. It lets you record information sources, how content was prepared and confirmed, and a document’s status, among other details. This is useful when AI agents create or update some of the content.
The type field remains the only field that is always required. It identifies the kind of information being described. Additional fields and their dependencies must follow the specification. Minimum format compliance does not mean that a document contains everything its reader needs.
Source: the current OKF 0.2 specification
Notice about the move to a separate project repository
How does OKF help organise knowledge for AI?
One document describes one concept or topic. Links connect it to other documents. A metric description can therefore point to data definitions, and instructions can point to the conditions they cover. Files can be reviewed, versioned and moved between systems.
The project repository also illustrates a separation of roles: one tool can prepare knowledge and another can read it. A file library can serve an agent, an internal search engine or a documentation browser. Each implementation still needs to provide access to the content and format support on the receiving side.
Source: the Open Knowledge Format repository
Conceptual diagram: source materials are verified, added to an organised knowledge base and can be read by a connected AI system. Original illustration based on OKF documentation, created using AI.
Semly’s knowledge base and OKF
Semly has a knowledge base and uses Open Knowledge Format. This is an important part of working with company information: both the answer itself and the organisation of the knowledge behind it deserve attention.
For the person responsible for marketing, this means a concrete task. Gather confirmed information about the offering, explain terms and remove contradictions. Use that material as the foundation for further descriptions, guides and answers to customer questions.
If you run a service business, start with the service scope, requirements and exclusions. If you sell products, organise use cases, specifications and compatibility. In both cases, readers should understand when the information applies.
The knowledge base and blog can serve different roles. The knowledge base maintains a precise answer to a specific question. A blog article helps readers understand a problem, compare approaches and decide what to do next. There is no need to duplicate the entire set of instructions in both places.
See also: how to divide topics between a blog and a knowledge base
Example: what does an “active account” actually mean?
Suppose a team wants to use AI to explain product reports. In one document, an “active account” means any account with valid access. In another, it means an account on which a specific action was performed. Without clarification, these descriptions can produce different results.
The sample data explains the method and does not represent any brand’s results. In this example, 80 accounts have valid access. At least one approved operation was performed on 28 of them during the week being analysed. If this is our definition of activity, the share of active accounts is 28 ÷ 80 × 100% = 35%.
The knowledge base should therefore describe the definition, the event that qualifies an account for inclusion, the period and the exclusions. The statement “activity is 35%” is not enough. It does not tell the reader what was counted or whether this value can be compared with another report.
| Knowledge element | Description in the sample data | Why record it? |
|---|---|---|
| Definition | An account with at least one approved operation during the week | Distinguishes access to the product from using it |
| Calculation basis | 80 accounts with valid access; 28 active | Makes the 35% result reproducible |
| Exclusions | Test accounts excluded from both groups | Prevents different populations from being mixed |
| Source and review | An identified event definition and the person responsible for checking it | Makes explanation and later updates easier |
The lesson is simple: agree on the definition first, then assess the AI response. If the system confuses an active account with an account that has access, check whether it received the right description. Changing the file format will not fix an unresolved definition.
What should you prepare in your company’s knowledge base?
Start with one area where misunderstandings hinder customer decisions or the team’s work. This might be the scope of implementation, a condition for using a feature or the meaning of a metric. You do not need to rebuild all your documentation at once.
Editorial checklist for preparing content: meaning, scope, source, currency, exceptions and relationships. This is a suggested way of working with information, not a list of required OKF fields. Original illustration created using AI.
- Choose a question that should receive an unambiguous answer.
- Write a short definition and the conditions under which it applies.
- Identify the source of the fact and the person responsible for confirming it.
- Describe exceptions and link the topic to the relevant instructions.
- Remove conflicting versions of the information from the materials you maintain.
- Check the system’s response to the basic question and to a boundary case.
A boundary case is particularly useful. If a service covers a certain type of device, also ask about an unsupported model. Good documentation should help identify the limitation, rather than encourage the response to extend the offering.
Does OKF improve visibility in Google and AI answers?
OKF concerns the way knowledge is recorded and exchanged. Visibility in public AI answers also depends on whether a system can access the material, considers it useful and uses it for a particular question. Implementing the format should not be treated as confirmation that a brand will be recommended.
Google explains that SEO fundamentals apply to AI Overviews and AI Mode. A page must be indexed and eligible to appear with a snippet. No special AI files or additional type of structured data is required. Meeting these requirements does not guarantee inclusion.
Source: Google Search Central - AI features and your website
Separate the two tasks. First, check that your knowledge base contains correct, current answers. Then assess how the company is presented in public AI systems. This helps distinguish a content problem from a situation where the material was not used.
Where should you start?
Choose one fact about your offering that is easy to misunderstand. Record its meaning, conditions and source. Check that the same rules apply in the knowledge base, on the product page and in the instructions.
This work provides a practical starting point for using OKF. Semly has a knowledge base and uses the format, but information quality still starts with facts. The easier it is to establish what a statement means and when it is current, the easier it is to check and use responsibly.
Technical sources were checked on 13 October 2026. The description of changes to the standard concerns OKF 0.2; it does not specify the version or scope of Semly’s OKF implementation.
Share:
