Google published something in June that reads like plumbing and matters like strategy. The Open Knowledge Format, or OKF, is a way to represent knowledge as a directory of markdown files with a thin layer of structured fields on top. It was built for companies to share internal data with AI agents. The interesting question is what happens when you point it at a public website.
Search Engine Journal asked exactly that, and the answer reframes how visibility works when the reader is a machine.
Pages Versus Graphs
Most machine-readable efforts so far have been flat. You take your pages, strip them to clean text, and hand the model a pile of documents. Each page stands alone. The model sees the words but not how the words relate.
OKF does something different. Each concept gets its own markdown file with a few queryable fields, type, title, description, tags, a timestamp, and a body. Then ordinary markdown links connect the files to each other. The result is not a stack of pages. It is a graph of relationships a machine can walk.
That distinction is the whole point. A flat copy tells a model what each page says. A graph tells it that one page is the framework underneath another, and a third is the narrower goal beside it. The machine stops reading isolated facts and starts reading structure, which is closer to how understanding actually works.
Why This Is a Visibility Question, Not a Technical One
Here is the part marketers should sit with. AI systems increasingly decide what your brand means by reading how your content connects, not just what each page claims. If your site is a set of disconnected pages that each assert you are the leader in your category, the model has words with no scaffolding. If your content shows how your ideas relate, the framework, the method, the proof, the outcome, the model can reconstruct a coherent picture of what you do and why it holds.
From building GEOflux, this matches what I see in the data. Brands that get represented accurately in AI answers tend to have content that hangs together, consistent claims that reference each other across pages. Brands that get summarized into a vague, wrong description usually have the opposite, a pile of pages that never link into an argument. The model is reconstructing a structure. Give it one worth reconstructing.
Practically, that means auditing your site for orphan pages, the ones that assert something important, link to nothing, and are linked from nothing. To a human skimming, an orphan page is fine. To a machine mapping relationships, it is a fact floating in space with no context, easy to misread and easy to ignore. Connect it to the pages that explain and support it, and you hand the model the scaffolding it needs to place you correctly.
This is a more demanding version of a point I keep making. I wrote about the llms.txt file almost nobody reads, where the lesson was that a shortcut file does not make you legible to AI. OKF is the non-shortcut. It asks you to actually structure your knowledge, which is harder and is the reason it might matter.
The Catch, and What to Do Anyway
OKF is not free to run. The moment your website changes, the bundle is wrong until you update it too. That synchronization cost is real, and it is the same tax every machine-readable format has quietly carried. A structured mirror of your site is only useful while it stays in sync with the site.
So I would not rush to publish an OKF bundle this week. Adoption is early, Google built it for internal use, and no major AI engine has committed to reading it from public sites yet. Chasing an unadopted format is how teams waste a quarter.
The move is to take the idea, not the file. You do not need OKF to make your site legible to machines. You need your content to have the structure OKF makes explicit. Concepts that link to related concepts. A clear frame that shows which page is the foundation and which is the detail. Consistent naming so the same idea is called the same thing everywhere, which lets a model connect the dots without guessing.
That work pays off no matter which format wins, because it is the substance the format is trying to expose. I argued that AI search will not cite you for engineering the appearance of authority. The flip side is that it will represent you well when your actual knowledge is structured well. OKF is one attempt to make that structure readable. The structure itself is the asset.
The web spent two decades optimized for a human skimming a page and a crawler counting keywords. The next reader is an agent trying to understand how your ideas connect. You do not have to adopt Google's format to serve that reader. You do have to stop publishing disconnected pages and start building something a machine can actually follow. Structure is becoming the visibility layer. Build it before your competitors make theirs legible first.