I Spent 3 Months Testing Knowledge Systems for AI—Here's the Free One That Works (18 Prompts + Organization Framework)

I have a confession to make: I am generally bad at notes.

I was bad at notes before AI, and AI is finally doing what AI is SUPPOSED to do—help me with my weak spots.

What I’m sharing here is from really bloody experience. I’ve tried a LOT of note-taking apps, and I’ve abandoned most of them. Evernote? Tried it. Obsidian? Tried that too. Notes? I have hundreds of notes buried by Apple’s terrible AI.

I’m finally ready to share how I approach notes in the age of AI.

I wanted to take my time with this piece, because notes are VERY personal and I wanted to write a guide that would feel relevant for YOU, not just for me.

After a ton of testing, I’ve come to a conclusion that surprised even me: for most people, for most things, NotebookLM is probably the correct notes solution.

Why? Three simple reasons:

1. It has a generous free tier and and is very easy to use (don’t sleep on that value)

2. Most actual value in notes is accurate retrieval around a theme

3. Google has built an insanely accurate retrieval model into NotebookLM

TLDR; this is a free tool that’s insanely accurate at recalling your notes so you don’t lose them. And that’s good enough for most people.

This begs the question: WHY is NotebookLM not the top choice on every internet listicle? Most of them will go for classic note-taking apps with lots of ‘features.’

That’s the issue: they’re too complex, and they’re too opinionated. The features lock you into a usage pattern, or they’re flexible but technical (like Obsidian).

NotebookLM is neither, which means you can easily customize it to you.

But the problem is because NotebookLM isn’t on the notes listicles (I KNOW) no one is really writing guides for it as a knowledge base.

Which means I couldn’t find a great guide for building a comprehensive notes system using NotebookLM, including multiple setup options, suggested prompts, how to use it with ChatGPT, etc.

So I fixed that. Here’s what you get:

18 copy-ready retrieval prompts organized by use case

• Research and learning (multi-source explanations, pattern finding, gap identification, evolution over time)

• Client and project work (background extraction, requirements by category, progress tracking, risk identification)

• Document comparison (contradiction finding, version diffs, completeness checks)

• Source verification (citation audits, provenance, exact quote extraction)

3 complete end-to-end workflow examples

• Learning a technical topic (retrieval in NotebookLM → design decisions in Claude)

• Client analysis (concern extraction with dates → alignment analysis → steering committee prep)

• Competitive research (strategy extraction per competitor → positioning gaps → differentiation in Claude)

4 organization schemas by role

• Consultant, Researcher, Product Manager, Founder

• Each includes typical active projects, archive patterns, and query cadence at scale

A 19 minute NotebookLM podcast about this article :)

Project management frameworks that prevent the common failure modes

• Naming conventions that scale to 20+ projects

• Decision framework for when to split vs. merge projects

• Size guidance (ideal: 15-50 sources, split at 75+, quality degrades at 100+)

• Archiving strategy and quarterly maintenance

Good vs. bad prompt comparisons

• Side-by-side examples showing why vague prompts fail

• How to frame for retrieval vs. synthesis

• Handoff patterns for moving from NotebookLM to Claude with context intact

The 6 most common pitfalls with specific fixes

• Asking for thinking instead of retrieval

• Project bloat from indiscriminate source adding

• Not copying extractions (chats don’t save)

• Vague prompts that return noise

• Ignoring citations (defeats the entire purpose)

• Using NotebookLM for final outputs (wrong tool for that job)

5-minute setup walkthrough

• Narrow project naming

• Source addition patterns (files, URLs, docs, transcripts)

• Scoping questions to inventory what you have

• First retrieval prompt to validate the system

Why did I write all this out? Because it’s HARD to do notes well.

It generally sucks. We’re afraid to lose notes. We’re afraid to get organized. We want AI to help but not hallucinate.

And I want it to be easy. NotebookLM can help with all this, but only if you know how to get organized. That’s what this guide solves for.

Who is this for? The list is long!

• Anyone who wants to not lose their notes again!

• Product Managers keeping notes on customers

• Engineers reviewing technical architecture conversations

• UX doing customer research

• Marketing doing competitive research

• Sales trying to track and learn from prospect conversations

• Students

• Vibe coders figuring out how to keep their build notes organized

• Professors, teachers, and academics

• Anyone trying to learn AI (I use it for this one!)

• People using OTHER notes apps (YES)

The list goes on. You get the idea. Happy note-taking!

One small note on why I think people using other notes apps would find this guide helpful: I’m framing it specifically in terms of how you retrieve clean context out of notes applications to hand to a reasoner LLM for thinking. I find most built-in AI in notes apps is not great (unless you do extensive tinkering), so the work done around prompts and workflows can cross-pollinate for you!

Alright, let’s dig in and start organizing our notes for AI, with AI, and most of all in ways that makes sense for OUR brains :)