A shared knowledge base written in Google's Open Knowledge Format can give your team's AI tools the context they usually lack: what you sell, to whom, in what words, and with what proof. You do not need an engineering project to start. A folder of well-linked markdown files is enough.
Where the idea comes from
Google Cloud published the Open Knowledge Format (OKF) in June 2026 as a way to describe a body of knowledge as markdown files with a small block of YAML frontmatter, readable by people and AI agents alike (Search Engine Journal covered the launch). Shortly afterwards, SEO consultant Marie Haynes published a walkthrough of her own "OKF brain": a personal, file-based knowledge store that gives AI agents structured context for her SEO work. As summarised by Let's Data Science, it uses typed pages such as concepts, entities, playbooks, references and system pages, plus a top-level index that an agent reads first.
Her example is personal. The same approach works well for a team.
Related: Vibe Coding a CRM: When to Build and When to Buy
Why marketing teams need this
Most B2B marketing teams already use AI for drafts, research and analysis. The results are often generic because the model does not know your business. People fix that by pasting the same background into every prompt, and each person pastes a slightly different version.
A shared knowledge base solves three problems at once:
- Consistency: every tool and every person works from the same positioning and facts.
- Speed: new hires and agencies get up to speed from one place.
- Accuracy: when a fact changes, you change it once.
Choose your page types
Keep the list of types short so the structure stays obvious. A B2B marketing team might use:
- Index: a single entry point that explains the folder and links to everything else.
- Concepts: the problems you solve and the category terms you use.
- Entities: your company, products, competitors and ideal customer profiles.
- Playbooks: repeatable processes such as launching a campaign or writing a case study.
- References: brand voice, approved claims, pricing rules and legal constraints.
Each file gets frontmatter with a type, title, one-sentence description, tags and a last-reviewed date.
Related: Build the Dashboard First, Then the AI Agent
Write pages that agents can use
A knowledge base only helps if the pages are specific. Good pages share a few habits:
- Lead with a one-paragraph definition in plain language.
- State facts once and link to the page that owns them instead of repeating them.
- Include what not to say, such as retired claims or banned phrases.
- Link related pages explicitly so the graph reflects how ideas connect.
- Add a source or owner for every fact that could change.
A page on your ideal customer profile, for example, should link to the concepts it cares about, the products that serve it and the playbooks used to reach it.
Connect it to your tools
How you use the folder depends on your stack. Common options include adding the relevant pages as context in your AI assistant's project or workspace settings, pointing coding and research agents at the folder, or storing it in a shared repository so changes are reviewed. Start with whichever tool your team uses daily and expand once it proves useful.
Keep it current
The biggest risk is decay. A knowledge base that contradicts your website is worse than none. Build maintenance into existing routines:
- Give each page an owner.
- Review pricing, product and positioning pages whenever those change.
- Run a quick monthly check for pages past their review date.
- Delete pages that no longer apply rather than leaving them to confuse agents.
Feed it with real buyer data
The most valuable knowledge for a marketing team is what real buyers do. Add pages that capture patterns from your own data: which industries visit most, which pages they read before a demo, which messages earn replies. Website visitor identification from Happierleads is one source for this, because it shows which companies and people are researching you, not just how many sessions you had. Summarise the patterns into the knowledge base so every AI draft reflects actual buyer behaviour.
An example page structure
To make this concrete, here is what a single ideal customer profile page might contain for a B2B software company. The frontmatter would declare it as an entity, give it a short title such as "Mid-market SaaS marketing leaders", a one-sentence description and a last-reviewed date. The body would then cover:
- Who they are: typical titles, company size range and the teams they manage.
- What they are trying to achieve this year and what usually blocks them.
- The words they use to describe the problem, taken from sales calls and support tickets.
- Which of your products or features they care about most, linked to those product pages.
- Objections that come up in deals and the answers that have worked.
- Links to the playbooks used to reach them and any public case studies that match.
Pages like this are useful to people as well as agents. A new sales rep can read it before a first call, and an AI assistant drafting an email can pull the right language and proof points without anyone pasting background into the prompt.
Treat the first month as a pilot. Pick two or three recurring tasks, such as drafting campaign briefs or writing follow-up emails, and compare output quality with and without the knowledge base as context. That comparison tells you quickly whether the effort is worth expanding.
The takeaway
Start small: an index, five to ten concept and entity pages, and your brand voice rules. Use it in one tool for a month and see whether drafts get better and faster. If they do, expand it and put maintenance on the calendar. The format is simple on purpose; the value comes from the discipline of keeping it true.
