Your AI isn’t underperforming. Your knowledge is under-organized.
When leaders tell me “AI feels cool, but nothing sticks”, my first question is never about the model.
It’s about the knowledge base.
AI can’t use what you haven’t organized.
In this week’s episode of ‘AI Literacy for Entrepreneurs’, I share the simplest system I know to make AI actually useful inside a small team: a four-folder knowledge base that pairs rules (how your AI behaves) with knowledge (what it must know).
It’s light.
It’s teachable.
And it’s the difference between one-off prompts and repeatable performance.
Rules vs knowledge (and why both matter)
Rules are your “how”: tone, guardrails, steps, order of operations. Knowledge is your “what”: offers, pricing, voice, proof, policies.
If you only give rules, you get generic copy.
If you only dump knowledge, you get rambling and mis-prioritized outputs.
The win lives in the pairing.
The four folders (copy this exactly)
Brand Voice - 2-3 of your best pieces, a do/don’t list (structures to avoid, emoji rules), and a one-page style card (headlines, Canadian spellings).
Product Facts - one-pagers for each offer (who it’s for, outcomes, components, timelines), proof points (quotes/case stats), and your positioning statement.
Policies and Pricing - current price ranges with inclusions/exclusions, payment terms/guarantees/refunds, and availability/process (booking → onboarding → delivery).
Gold Standard Examples - gold-standard work (before/after if you have it), templates/checklists, and boilerplate (bio, short company description, 3-line pitch).
This is where hallucination drops and first-pass quality jumps.
Freshness beats “set and forget”
Knowledge decays.
Offers change.
Prices move.
Put a freshness rhythm in your calendar (monthly or every two months). Version by date, keep a one-line changelog, and you’ll maintain quality without heroic rewrites. Twenty minutes is enough if you keep it lean.
Safety isn’t optional
Never paste contact information, confidential contracts, or unreleased IP into public LLMs. Anonymize sensitive examples, summarize legal docs instead of dumping full text, and set access so everyone can see the knowledge base but only a few can edit. Responsible adoption is a competitive advantage.
A live example (and why this works)
I used this exact setup to produce Northlight landing pages: audience, tone, and format in the rules; voice samples, offer one-pager, and past winners in the knowledge. The draft came back ~80% right. Ten minutes of edits later, we shipped. That’s what AI that compounds feels like.
Your 48-hour challenge
Create the four folders.
Drop 1-2 docs into each (rough is fine).
Test on one real deliverable and time it.
If you want eyes on your setup, bring your folder map to my MPC open house. We’ll spot gaps, suggest missing docs, and tune it live. DM me “Open House” on LinkedIn and I’ll share the details.
For much more detail, listen to the companion episode to this article - here’s to EP 247 of ‘AI Literacy for Entrepreneurs’
Design the knowledge.
Then let AI do the heavy lifting.
