Steps to build a starter AI thinking system

- Account access to NotebookLM, Gemini, and PerplexityActiveCampaign account (with CRM access) - A folder of source material you can upload: proposals, frameworks, customer interview transcripts, past campaigns, decision rules - One sentence describing what your system needs to answer six months from now - An hour of uninterrupted time to draft the objective and pull source material

- Write the forcing question: Answer this in one sentence: "What is the single most important thing my system should be able to answer six months from now?" - Pressure-test it: If the question doesn't determine which documents are worth uploading, it's too broad. Sharpen until it does. - Write it down somewhere visible: Pin it in your notes, your project folder, or the description field of your first notebook. Every later decision points back to it.

- Identify the three or four questions that matter most: These are the questions your system needs to be able to answer. Write them as bullet points under your forcing question.

- Build the standing library in NotebookLM: Upload canonical sources—books, foundational reports, frameworks—that will stay relevant across many projects. Treat this as a reference shelf, not a project workspace. - Spin up a focused second notebook for the live challenge: Reuse the canonical PDFs but add project-specific files. Cassady runs his system this way: one permanent shelf, one fresh workspace per challenge. - Use Gemini for synthesis against the standing library: Connect Gemini to your NotebookLM and ask it to synthesize across the canonical sources. This is where deeper patterns surface. - Use Perplexity for anything recent: Perplexity surfaces citations that other tools bury. Use it when the question depends on current data the standing library can't reach.

- Pull the proposals that won: Not all of them. The ones that closed. - Pull the frameworks your team actually uses: Decision rules, naming conventions, internal playbooks. If a framework lives only in someone's head, write it down before uploading. - Pull customer interview transcripts: Real voice beats summary. Raw transcripts produce sharper outputs than cleaned-up notes. - Pull past campaign artifacts: Briefs, post-mortems, performance data. Specificity is the goal. - Audit for AI-generated content: Anything that was drafted by AI and never humanized contaminates the library. Either rewrite it in your team's voice or leave it out.

- Run a bulk-generation pass against the objective: Cassady uses tools that produce hundreds of variations in about 15 minutes. The goal isn't to use every idea—it's to have enough options that the strongest one stands out.

- Cluster the output: Group similar ideas. The clusters tell you where the obvious answers live and where the unexplored territory is. - Pick the 3-5 strongest ideas to develop further: Apply SPARKS to each one—starting with S (speak it out) on what makes the idea worth developing.

- Run every output through SPARKS before it goes back into the system: Every artifact you save into the library should have been pressure-tested—S (speak), P (pivot), A (ask), R (reframe), K (keep asking), S (stop and think). - Keep your bullshit hat handy: Before you read any AI output, deliberately put on the skeptic's mindset. Without it, flattery looks like validation. - Review the library quarterly: Remove sources that no longer reflect your work. Add new proposals, transcripts, and frameworks as they accumulate. - Track what the system is good at: Note which questions it answers well and where it still falls short. The gaps tell you what to upload next.