A complete system for learning anything faster with AI — built from the same setup I run across 16 real notebooks. Non-technical, no code, roughly 30 minutes to set up.
You need three things you already have: a computer, a free Google account, and one topic you're trying to get better at.
1. The folder structure
Every topic gets its own folder. Not a shared "notes" dump — a dedicated folder per subject, because the next two parts of this system (your instructions file and your NotebookLM notebook) both point at that folder.
Create one parent folder, then one subfolder per topic:
AI Learning OS/
├── 01-Public-Speaking/
│ ├── sources/
│ ├── journal.md
│ └── CLAUDE.md
├── 02-Excel-for-Finance/
│ ├── sources/
│ ├── journal.md
│ └── CLAUDE.md
├── 03-Guitar/
│ ├── sources/
│ ├── journal.md
│ └── CLAUDE.md
Inside each topic folder:
- `sources/`: every PDF, saved article, and note you collect on this topic. This is what you'll upload into NotebookLM.
- `journal.md`: one running text file. Every takeaway, every "oh, that's why" moment, gets appended here. Never delete from it — it becomes a timeline of how your understanding changed.
- `CLAUDE.md`: your instructions file for this topic (built in section 2).
If you already have an AI assistant that reads project files (Claude Projects, ChatGPT Projects, or similar), point it at the topic folder. If not, you'll paste the instructions file's contents in manually each session — still works, just one extra copy-paste.
Start with 2-3 topics, not ten. A folder you never open is worse than no folder.
2. The master instructions file
This is the single most useful file in the system. It's a plain-text sheet that tells any AI assistant who you are, what you're learning, and how you like things explained — so you stop re-explaining yourself in every new chat.
Save this as CLAUDE.md (or instructions.txt, the name doesn't matter) inside each topic folder, or one shared version at the top level if your topics overlap.
Copy this template and fill in the blanks:
WHO I AM
- Name / role: [e.g. "Noam, marketing manager, no coding background"]
- Current level in this topic: [beginner / know the basics / rusty / advanced]
WHAT I'M LEARNING
- Topic: [e.g. "Public speaking for work presentations"]
- Why: [the real reason — "I have a board presentation in 6 weeks"]
- Success looks like: [one concrete outcome you'll be able to do]
HOW I LIKE EXPLANATIONS
- [ ] Short and blunt, skip the theory
- [ ] Analogies to things I already know: [list 2-3, e.g. "sports, cooking, sales"]
- [ ] Show me an example before the rule, not after
- [ ] Ask me questions instead of just telling me
- Words/jargon that lose me: [list them — the AI should define these, not assume]
CURRENT PROJECTS / CONTEXT
- What I'm working on right now related to this: [e.g. "drafting slide 1 of my Q3 deck"]
- What I already tried that didn't work: [saves you from getting the same bad advice twice]
WHEN YOU EXPLAIN SOMETHING TO ME
- Check I actually understood before moving on — ask me to explain it back in one sentence.
- If I ask the same type of question twice, that's a sign — go slower, not faster.
Fill it out once per topic, in five minutes, and every AI conversation about that topic starts already knowing you. No more "I'm a beginner, please explain simply" typed into every new chat window.
3. NotebookLM as the study layer
Your topic folder holds the raw material. NotebookLM turns it into something you can actually absorb — including a podcast version, hands-free.
Set up one notebook per topic folder:
1. Go to notebooklm.google.com, sign in with your Google account (free — no card).
2. Click New notebook, name it to match your folder (e.g. "Public Speaking").
3. Upload sources from your topic's sources/ folder — PDFs, saved articles, your notes.
The source recipe that works: aim for roughly 5 videos plus whatever web search turns up, landing around 15-20 sources per notebook total.
- 5 videos: search YouTube for your topic, grab the URLs of the 5 best explainer/tutorial videos, add each as a source (NotebookLM accepts YouTube links directly — no download needed).
- Web search finds: as you research, save the best articles, guides, and PDFs you come across straight into sources/, then upload them too.
- Don't overthink source quality on the first pass — you can always add more later, up to 50 per notebook on the free plan.
Turn any single PDF into a podcast (the fastest single move in this whole system):
1. Open the notebook, click Add source, upload the PDF.
2. Click Audio Overview in the Studio panel, then Generate.
3. Wait a few minutes — two AI hosts build a ~10-minute conversation out of the source.
4. Press play. Listen like a podcast — on a walk, during a commute, while cooking.
Free-plan limits worth knowing: 3 Audio Overview generations per day, 50 sources per notebook, 100 notebooks total. None of that will bottleneck a normal learning habit — three podcasts a day is more than most people need.
4. The weekly loop
A notebook without a rhythm goes stale. Run this once a week, per active topic, about 20 minutes total:
1. Collect. Drop anything new you found this week — an article, a video link, a PDF — into sources/ and add it to the notebook.
2. Generate the podcast. Click Audio Overview → Generate. Let it build.
3. Listen on a walk. No screen, no multitasking. Just the two hosts and your topic.
4. Ask 3 follow-up questions. Back at the notebook, type into the chat the 3 things that weren't fully clear, or that you want to go deeper on. This is where the real learning happens — the podcast is the setup, the follow-up questions are the payoff.
5. Journal the takeaway. Open journal.md and write 2-3 sentences: what clicked this week, what's still fuzzy, what you're doing differently now. Takes two minutes, compounds over months.
That five-step loop is the entire system running. Everything else in this guide exists to feed it.
5. Bonus: the journal-analysis trick
Once you've been running the weekly loop for a few weeks, your journal.md becomes a source worth analyzing on its own — not for the topic, but for you.
How to run it:
1. Copy the contents of journal.md (or export it as a PDF/text file).
2. Upload it as a source into a notebook — a dedicated one, separate from your topic notebooks.
3. Generate an Audio Overview.
4. Listen to two AI hosts discuss your own writing back to you — patterns, contradictions, things you said in week 2 and again in week 6 without noticing.
It's an odd feeling the first time. It's also the fastest way to catch a blind spot you're too close to see in your own words.
Five prompts to steer the hosts when you're in the notebook chat (not the audio — type these after listening):
- "What's one thing I said I'd do in an early entry that I never actually did?"
- "What topic or frustration comes up more than once — and did I ever resolve it?"
- "If you had to guess what I'm avoiding based on this journal, what would it be?"
- "Compare how I talked about [specific project/goal] in the first half of these entries versus the second half — what changed?"
- "What question should I be asking myself that I haven't asked yet?"
Run this monthly, not weekly — you need enough entries for a pattern to actually show up. A single week of journaling won't give the hosts much to work with; four to six weeks will.
That's the whole system
A folder per topic, one instructions file so the AI already knows you, a notebook that turns your sources into a podcast, a weekly loop that keeps it alive, and a once-a-month mirror pointed at your own journal. Thirty minutes to set up, and it runs itself from there.
Want the fast version first? Read the companion how-to, Turn Any PDF Into a Podcast — Free (NotebookLM Audio Overview), on this site.