Google's Open Knowledge Format (OKF) was designed to help AI agents understand a company's internal data, but it may also be a cheap way to explain your public website to machines. For most B2B teams the right move today is a small, low-risk experiment, not a rebuild of the site.
What Google actually released
In June 2026 Google Cloud published the Open Knowledge Format, a specification for describing a body of knowledge as a folder of markdown files. Each file carries a short block of YAML frontmatter with a handful of queryable fields such as type, title, description, resource, tags and timestamp, and the body explains the concept in plain language. Search Engine Journal's coverage of the announcement describes the intended users as documentation teams, data specialists and engineers producing knowledge, with AI agents, LLMs and search indexes consuming it. The specification itself lives on GitHub.
The target was never marketing websites. OKF was meant for things like datasets, metrics, tables and APIs scattered across a company's internal systems.
Related: Unlocking Growth: How Google Analytics Can Transform Your Website's Performance
The idea of using OKF on a public site
Slobodan Manic, host of the No Hacks podcast, argued in a piece for Search Engine Journal that the same format could help public websites. His point: an AI agent visiting a site today reads pages one at a time and sees a flat copy of each. It does not see how your ideas relate. An OKF bundle, where each concept gets its own file and files link to each other, gives the agent a map instead of a pile of pages.
He tested it on his own site with a small set of files covering his brand, himself as host, and the frameworks he teaches. He was also candid about the limits: the bundle is a second copy of your knowledge that goes stale when the site changes, Google built the format for internal data, and no agent is known to read website knowledge bundles today.
Why this matters for B2B marketers
B2B buyers increasingly ask AI assistants to shortlist vendors, compare features and explain categories before they ever land on a website. When that happens, the assistant builds its answer from whatever it can piece together about you. Three problems show up again and again:
- Category confusion: the assistant files you under the wrong category or compares you to the wrong competitors.
- Stale facts: old pricing, retired features or a previous positioning keep resurfacing.
- Missing relationships: the assistant knows your product name but not which problem it solves for which buyer.
A structured, linked description of your core concepts is one way to reduce those errors, whether an agent ever reads an OKF bundle directly or simply benefits from the cleaner site architecture you build while writing one.
Related: The Ultimate Guide to Open Web Analytics for B2B Growth (2026)
What an OKF bundle for a B2B site could contain
Keep the first version small. Eight to fifteen files is plenty to learn whether the exercise is useful. A sensible starting set:
- Company: who you are, who you serve, where you operate.
- Product: one file per product or module, with the job it does.
- Problems: one file per buyer problem you solve, linked to the products that solve it.
- Audiences: one file per ideal customer profile or persona.
- Concepts: the terms your category uses, defined the way you use them.
- Proof: links to documentation, help articles and public case studies you can stand behind.
Each file gets a clear type, a one-sentence description and a canonical URL on your site, so the bundle points back to the pages that should rank and be cited.
The honest costs
The strongest argument against OKF on a website is maintenance. Every time pricing, packaging or positioning changes, the bundle has to change too, or you are publishing contradictions. Teams that already struggle to keep their help centre in sync with the product should be wary of adding another source of truth.
There is also no guaranteed payoff. Unlike robots.txt or sitemaps, nobody has said their crawler consumes website knowledge bundles. You would be building ahead of adoption.
A practical way to test it
If you want to experiment, treat it as a content architecture exercise with an optional machine-readable output:
- Write the concept files for your own team first and use them internally, for example as context for the AI tools your marketers and sales reps already use.
- Link each file to its canonical page and fix any page that contradicts the file.
- Publish the bundle only once it is accurate, and add its review to the same checklist you use for pricing and product changes.
- Re-check in a quarter whether any major assistant or crawler has announced support.
The internal use case pays off even if no external agent ever reads the files. A clean, shared definition of what you sell and to whom improves sales decks, onboarding and ad copy as well.
Where visitor intelligence fits
Structured knowledge helps machines describe you correctly. It does not tell you which companies and people are actually researching you after an assistant sends them your way. That is the gap website visitor identification fills: Happierleads identifies the person and company behind a B2B website visit, so the traffic that AI-assisted research sends you becomes a list of named prospects rather than an anonymous session count.
Mistakes to avoid if you try it
Early experiments with any new format tend to go wrong in the same few ways. Watch for these before you publish anything:
- Copying pages into files. A bundle that simply repeats your page copy adds nothing. Each file should define one concept and explain how it links to others.
- Marketing language. Agents and people both need plain definitions. "The leading platform for growth" tells a machine nothing about what you do.
- Contradicting the site. If the bundle says one thing about pricing and the pricing page says another, you have made the stale-facts problem worse.
- No owner. Someone must be responsible for reviewing the files when products or positioning change.
- Expecting traffic. Nothing announced so far suggests a bundle will bring visitors or rankings on its own. Judge it on internal usefulness first.
A good sanity check is to hand the bundle to a new colleague and ask them to explain your business back to you. If they can do it accurately in five minutes, the structure is working. If they cannot, an AI agent will not either.
The takeaway
OKF is worth understanding and cheap to trial, but it is not an SEO lever yet. Build a small, accurate concept bundle for internal use, tie every file to a canonical page, and publish it only if you can keep it current. Meanwhile, keep investing in the basics that every crawler and assistant already reads: clear pages, consistent facts and a site that makes your category and buyer obvious.
