Steps to build a Living Knowledge system
This checklist is based on the You don’t have an AI problem. You have a “Dead Knowledge” problem featuring Bryan Cassady's Living Knowledge system, published on the AI Lab by ActiveCampaign.
Get the checklist
By the end of this checklist, you’ll have a Living Knowledge system: an interview-grounded True N.O.R.T.H. brief connected to a two-tier NotebookLM library that reviews your own proposals before a human does. It’s based on the workflow that Bryan Cassady, innovation professor and author of The Generative Organization AI Playbook, runs with marketing teams. One client used it to cut average proposal revisions from 30 to 3. The five steps take about a day to set up.
Before you start, make sure you have:
- Access to NotebookLM: notebooklm.google.com
- Access to Gemini: gemini.google.com/app
- A list of interview subjects: a mix of senior executives and front-line staff
- A way to collect written responses: the post doesn’t name a specific survey tool; any form or doc that lets you gather and export responses works
The workflow
Phase 1: Interview your people before touching documents
After this phase, you’ll have: interview transcripts uploaded to NotebookLM and a clear picture of where leadership and front-line staff disagree
- Send open-ended questions to a mix of senior executives and front-line staff: ask what the company does, its value proposition, the problem it actually solves, and what it’s not doing
- Upload interview transcripts to a NotebookLM notebook
- Query NotebookLM for agreement and disagreement:
Starter prompt (copy and edit):
What do respondents agree on about what this organization does and who it serves?
Follow-up prompts:
Where do responses contradict each other? Identify the three sharpest disagreements.
What phrases do people use to describe our value that we are not using in our official messaging?
“Senior leadership usually thinks the strategy is crystal clear,” Cassady says. “And the people closest to the customer are telling a completely different story. The gap between those two things is where every positioning problem lives.”
Phase 2: Build your True N.O.R.T.H.
After this phase, you’ll have: a one-page strategic anchor document to upload as the first source in your knowledge system
ChatGPT version, Gemini version
- Open the True N.O.R.T.H. GPT: ChatGPT version or Gemini version
- Run the starter prompt:
Use this
I would like your help writing a True N.O.R.T.H. for this challenge: [describe your project or campaign objective]. Ask me questions to build this step by step.
- Answer the GPT’s questions for each letter: Narrative (why this matters), Objective (the single-sentence win state), Restrictions (what’s off-limits), Tactical constraints (word counts, format, tone), Here’s the starting place (seed sources), plus the one-line “True” headline
- Save the one-page output as the first source in your knowledge system
Phase 3: Build the two-tier library (and keep the AI slop out)
After this phase, you’ll have: a standing library notebook and a live-challenge notebook in NotebookLM, seeded with vetted sources
- Create the standing library notebook: upload past campaign briefs, customer interview transcripts, frameworks the team actually uses, winning proposals (and a few losses with notes on why), post-project retrospectives, and the Step 1 interview transcripts. Upload entire folders, not individual files.
- Create a live-challenge notebook for each new project: reuse the standing library’s sources and add the current brief, client objectives, and competitive research
- Leave out AI-generated and sanitized content: skip LinkedIn posts, press releases, and approved messaging documents, since they’re written to sound good rather than be accurate
“Nine out of 10 LinkedIn posts from founders right now are written by AI,” Cassady says. “If you upload all of that, you’ve infected your system with something that sounds nothing like the actual human you’re trying to build around.”
Phase 4: Connect the library to AI
After this phase, you’ll have: Gemini running with live access to your NotebookLM knowledge system
- Open Gemini
- Click the plus sign in the lower left corner
- Select your NotebookLM notebook as a connected source
- Test it with a competitive-intelligence prompt:
Use this
I’ve uploaded [NUMBER] competitor campaigns into this notebook. Based strictly on what’s there, identify the three angles nobody is using. For each one, explain why it’s absent and whether that absence is a gap or a deliberate choice.
Phase 5: Generate in bulk, then let the system evaluate
After this phase, you’ll have: a batch of proposal or idea variations, pre-screened against your knowledge system before a senior reviewer sees them
- Generate variations against your True N.O.R.T.H.: Cassady generates hundreds of options in about 15 minutes so the right answer becomes obvious rather than chosen by default
- Run the evaluation prompt before anything reaches a client:
Use this
Before I share this with the client, evaluate it against everything you know about their preferences, their past feedback, and their stated objectives. What are the three weakest points? What contradicts something they’ve told us? What would make them push back immediately?
- Fix flagged contradictions and weak points before the deliverable goes to a senior reviewer
“The system didn’t replace their judgment,” Cassady says. “It made their judgment sharper and faster.”
Quick reference
- Total time: About a day for initial setup (interviews, True N.O.R.T.H., library build, connection); ongoing maintenance after that
- Tools needed: NotebookLM, Gemini, an interview/survey method [tool not specified in source]
- Key output: A Living Knowledge system that generates proposals and flags contradictions (like a client’s already-rejected recommendation) before a human reviewer has to catch it
Related
More data from the AI Lab.