Google Just Shipped Another AI File Format. Here's Why We're Not Rushing.
On 12 June 2026, Google Cloud published the Open Knowledge Format (OKF), a way to package knowledge as plain markdown files that AI agents can read. It is genuinely well designed. It is also the third "feed your site to AI" format the industry has been handed in eighteen months, after the AI crawler directive wave (GPTBot, ClaudeBot, CCBot) and llms.txt. Like the others, it is not read by the consumer AI engines that decide whether ChatGPT or Perplexity recommend your business. OKF was built for internal company knowledge, not public visibility. If you run a business that wants to be found in AI search, nothing about your to-do list changed this week.
Two days ago, Google Cloud released the Open Knowledge Format. Within hours, the takes arrived: the new standard for AI visibility, the thing you need to do now, and at least one free tool, built the same week, to convert your website into an OKF bundle.
We have seen this film before. So before you spend a weekend restructuring your site, here is what OKF actually is, what it is not, and why the honest answer for almost every business is "interesting, not urgent."
What OKF actually is
OKF is a specification for storing knowledge as a directory of markdown files, each one a "concept," a table, a metric, a playbook, an API, with a little structured data at the top of each file. Concepts link to each other with ordinary markdown links, which turns the folder into a graph an AI agent can walk. The entire v0.1 specification fits on a single page, and the only thing it strictly requires of each file is a "type" field.
It is deliberately minimal and, credit where due, elegant. It formalises a pattern that has been emerging organically for a while: the "LLM wiki," the CLAUDE.md and AGENTS.md convention files developers already use, the Obsidian-style vault pointed at a coding agent. Google took that pattern and wrote it down as an open, vendor-neutral standard.
Here is the part the breathless coverage skips: OKF was built to solve an internal problem. The scenario in Google's own announcement is an AI agent inside a company trying to write a SQL query and having to piece together context scattered across catalogues, wikis and code comments. OKF is the shared format that fixes that. It is plumbing for agents working inside your organisation, not a beacon for the AI assistants your customers use.
Why it doesn't change your AI visibility, yet
The consumer AI engines that decide whether your business gets recommended, ChatGPT, Gemini, Claude, Perplexity, Grok, do not read website-hosted OKF bundles today. There is no evidence any of them do. The format is two days old, it is explicitly version 0.1, and Google itself calls it a starting point with open questions still unanswered.
So if you build an OKF bundle for your website this week, the honest description of what you have done is: an experiment, and a small bet on where machine-readable formats might be heading. Not a tactic that earns you a single AI citation today. There is nothing wrong with placing that bet if you have spare time and like being early. There is everything wrong with being sold it as a must-do that will get you into ChatGPT, because that is not true.
We've been here twice already
This is the part we can speak to from experience rather than speculation.
The first "feed your site to AI" moment came in mid-2023: AI crawler directives in robots.txt. GPTBot, ClaudeBot, CCBot, PerplexityBot. Every SEO blog said configuring these rules was the key to being found by AI. In reality, allowing crawlers was necessary but not sufficient, and the coverage vastly overstated the impact of a few lines in robots.txt.
Then, eighteen months ago, the must-do format was llms.txt, a file you put at your domain root to tell AI systems what mattered on your site. It was going to be the key to AI visibility. We built llms.txt detection into our audit and, for a while, weighted it heavily.
Then, last month, Google published its official guide to optimising for AI features and put llms.txt on the list of things you do not need. We re-weighted it down the same week and said so publicly. No major AI platform has ever confirmed using it. It turned out to be a low-cost, optional, largely unproven file, exactly what we now believe OKF is for public visibility.
After OKF it will be something else again. The pattern is the constant: a new machine-readable format appears, a wave of "this changes everything" content follows, free tools spring up to generate it, and twelve months later the consumer engines still mostly ignore it while the fundamentals carry on quietly doing the work.
What actually decides whether AI recommends you
None of this has moved in eighteen months, through three format launches, because it is not about formats. In our audit of over 1,000 UK accountancy websites, 65% scored below 60 out of 100 on basic AI readability. The failures are not missing file formats. They are missing fundamentals.
AI engines recommend businesses they can clearly read and have reason to trust:
- Content that says exactly what you do, who for, and where. In plain text, specific enough for a machine to match to a real question.
- Named people with real credentials behind that content. Not an anonymous corporate voice.
- Genuine mentions on sources the AI already trusts. Earned, not manufactured.
- A site that is technically accessible. So the crawlers can actually read it in the first place.
- The same business, described consistently, everywhere it appears.
That list has survived the crawler directive wave, llms.txt, Google's AI guide, and now OKF. It will survive the next format too. If your fundamentals are right, a new file format is a marginal optimisation. If they are not, no file format will save you, which is the part the bandwagon never mentions.
So should you do anything about OKF?
For most businesses: no, not yet. Keep an eye on it. If the consumer engines start reading website OKF bundles, that changes the calculation, and we will say so, here, with the evidence, the way we did when Google contradicted llms.txt.
If you run a data or engineering team and you are drowning in scattered internal documentation, OKF is worth a serious look right now, for its actual purpose: making your internal knowledge portable and readable by the agents working inside your organisation. That is a real problem it genuinely solves.
But if you are a small business owner wondering whether OKF is the thing that finally gets you recommended by ChatGPT, the answer is the same as it was before Thursday. Get the fundamentals right, and measure where you actually stand.
One note on naming, since it trips people up: Google's Open Knowledge Format is unrelated to the Open Knowledge Foundation, a separate non-profit, and to a supply-chain spec that shares the same initials. When we say OKF, we mean Google Cloud's 2026 markdown format.
Want to know where your business actually stands?
Run a free SearchScore audit. Your AI Search, SEO and conversion scores across 250+ signals, in about 60 seconds, no email required.
See how AI visibility works for your field: We've built industry-specific guides for therapists, accountants, photographers and more. Browse all industries →