What to Put in Your Company Knowledge Base: The Complete Template

An AI assistant is only as good as what it's trained on. Trained on nothing, it guesses, and guessing is where AI systems earn their bad reputation — hallucinated pricing, made-up policies, confidently wrong answers. Trained on a real, current, well-organized knowledge base, the same system answers correctly, drafts in your voice, and knows when to say "let me get a human" instead of improvising. This template is the folder structure and document list we use to stand up that knowledge base for a business, in order of what matters most.

Most businesses already have 60–80% of this content — it's just scattered across old PDFs, a founder's head, group chat threads, and a website nobody's updated since a rebrand. The work isn't writing from scratch; it's consolidating, and then keeping it current.

How to use it: Work through the folders in order. For each document, mark Have, Needs update, or Missing. You don't need every document below on day one — Must-have items are what actually block an AI assistant from being useful; Nice-to-have items improve it over time.

What to exclude

• Anything you wouldn't want quoted back verbatim. An AI assistant will eventually surface exact phrasing from source documents — internal jokes, off-the-cuff Slack messages, and unreviewed drafts don't belong in the corpus.
• Outdated pricing or policy versions. Archive them separately. A knowledge base with three versions of a price sheet produces three different wrong answers.
• Legal or compliance language you haven't had reviewed. If a policy touches liability, refunds, or health/financial claims, have it checked before it becomes something an AI repeats to customers.
• Customer PII beyond what's needed for the workflow itself. Names and case details from past work should be scrubbed or generalized unless there's a specific, consented reason to keep them.

How to score your knowledge base

Count your must-have documents marked Have (present and current).

FAQ

How often does the knowledge base need to be updated? Whenever pricing, policy, or service offerings change — treat it like your website, not like a one-time document. Stale answers are the single biggest source of AI assistant complaints, and they're entirely preventable.

Does this work for a business with no written documentation at all? Yes, but expect the first pass to take longer — most of the work is capturing what's currently only in people's heads. Starting with services and pricing (Section 2) gets you to a usable minimum fastest.

Can we start an AI assistant with a partial knowledge base and add to it later? Yes, and it's the normal path. A partial but accurate knowledge base that says "let me check on that" for gaps is far better than a complete-looking one that guesses. Escalation rules (Section 6) are what make a partial knowledge base safe to launch with.

Put it to work

A knowledge base is only worth building if something puts it to use. Flon builds internal knowledge assistants trained on exactly this structure — docs, site, and past work — scoped after a Blueprint and operated monthly. Related reading: company knowledge base in the glossary, and the AI readiness assessment to check whether your data is deep enough yet.