How to Extract Any Expert's Knowledge Using AI

Tony Robbins built his entire career on one idea: if you want results, find people who already have those results, and model them. Study their patterns. Their routines. Their habits. Their thought process. Their worldview. How they actually think when nobody's watching. That's the shortcut to compressing a decade of trial and error into a few months of intentional learning.

For most of history, modeling someone meant either knowing them personally or spending hundreds of hours consuming everything they'd put out into the world. Books. Interviews. Lectures. Speeches. The cost of getting inside someone's head was massive, which is exactly why so few people actually did it.

That bottleneck is gone.

The amount of high-quality information sitting on YouTube alone is absurd. Founders sharing how they actually built their companies. Investors walking through their decision frameworks. Doctors explaining cutting-edge protocols. Operators showing real numbers from real businesses. Twenty years ago you'd have paid five figures to sit in a room with these people. Now they're publishing it for free. Same goes for podcasts, blog archives, long-form interviews, AMAs, and conference talks scattered across the web.

The only thing standing between you and modeling the smartest people in any field is the time cost of consuming their work. And that's the part AI just collapsed.

This is exactly what AI should be doing for everyday people. Not generic chatbot conversations. Not productivity hacks that save you 5 minutes a day. Compressing hours, days, or weeks of expert content into the actual essence, so you can study what makes people successful and apply it to your own life. Not the surface-level stuff. The real signal.

I run this protocol every week. It's how I study creators, founders, and thinkers I want to model, and it's replaced what used to be two weeks of work per person with about an afternoon. Sometimes 30 minutes if I just need a specific answer.

The setup uses YouTube and NotebookLM because that's the simplest entry point. The same protocol works for podcast feeds, blog archives, book chapters, and any source NotebookLM can ingest (PDFs, Google Docs, web articles, audio files). YouTube is the on-ramp. The thinking layer is the same regardless of source.

This isn't a tutorial. Plenty of those exist. This is the actual protocol I run, including the meta-moves nobody else is teaching: the four-pass method, the comparative analysis stack, the contradiction-hunting prompts, the evolution tracker, and the storage layer that makes the work compound.

If you've already seen the basic "load YouTube into NotebookLM" guide somewhere, that's the first 10% of what's possible. The rest is below.

What you'll need

• Chrome browser (or Chromium-based: Brave, Edge, Arc)
• YouTube to NotebookLM (free Chrome extension)
• A Google account with NotebookLM access (free at notebooklm.google.com)
• A note-taking system you actually use (Notion, Obsidian, Apple Notes, anything)

Stack is free. The note-taking system matters for the storage layer at the end.

The 5-minute setup

1. Install the extension. Go to the Chrome Web Store." Install the one with 300K+ users (developer: yt.to.notebooklm). Pin it to your toolbar.
2. Navigate to the creator's channel. Open YouTube. Go to the channel you want to study. Click into their "Videos" tab so you see their full library.
3. Click the extension button. With the channel page open, click the extension icon. You'll see options to send the whole channel, a specific playlist, or just the videos currently visible. Pick "send channel."
4. Choose your notebook. A prompt appears asking whether to create a new notebook or use an existing one. Pick "create new" and name it after the creator. The extension imports the channel's videos directly into NotebookLM as sources.
5. Wait for transcription. NotebookLM transcribes and indexes everything in the background. Takes a few minutes depending on volume.

Done. The notebook is now a queryable conversation with their entire body of work.

A note on source limits

NotebookLM's free tier caps each notebook at 50 sources. That's a hard limit. For most creators that's plenty. For prolific channels with hundreds of videos, you have three options:

• Prioritize within free tier: Load 25 most recent (their current thinking) + 25 most-viewed (their best material). The extension lets you pick which videos to import. This covers 95% of what matters for most creators.
• Split into multiple notebooks: "Creator X — early work" and "Creator X — recent work." You lose cross-source querying within a single notebook, but you can run the same prompts across both and compare.
• Upgrade to NotebookLM Pro: $19.99/month bundled with Google AI Pro. Raises the cap to 300 sources per notebook. Worth it only if you're doing this constantly across creators with massive libraries.

I run Free for most research and Pro when I'm going deep on someone for a specific decision.

The four-pass research method

This is the protocol I run every time. Each pass surfaces something different. Run them in order. Total time: 45 to 90 minutes depending on how deep you want to go.

Pass 1: The worldview pass (10 minutes). What does this person actually believe? Where do they disagree with the consensus in their field? This is the layer most people miss. Surface beliefs are easy to find. Contrarian beliefs are where the real signal lives.

Pass 2: The frameworks pass (15 minutes). What repeating models, frameworks, and mental shortcuts do they use? Most strong thinkers have 3 to 7 frameworks they apply across topics. Find those frameworks and you understand how they reason.

Pass 3: The strategy pass (20 minutes). How do they actually run their business or career? What's their content strategy, growth strategy, monetization model, positioning? This is where you separate the talkers from the operators.

Pass 4: The synthesis pass (15 minutes). Put it together. What is this person fundamentally about? What are they wrong about? What's the one thing you'd take from them?

By the end of these four passes, you have a more thorough understanding of someone than 99% of their actual audience. Most fans don't ever sit and do this work.

The twelve prompts, organized by pass

These are the prompts I actually use, grouped by which pass to run them on.

Pass 1: Worldview

Strongest opinions:

What are this person's strongest opinions? Be specific. What do they believe that most people in their space would disagree with? Quote them directly when possible.

Hidden contrarian takes:

What does this person believe that contradicts mainstream advice in their industry? Where are their views genuinely unusual versus where are they just repackaging consensus?

Belief evolution:

Has this person changed their mind on anything important across their content? Show me where their thinking has evolved over time and what triggered the shift.

Pass 2: Frameworks

Frameworks extraction:

What repeating frameworks, models, or mental models does this person use across their content? List each one with a brief explanation, the videos where it appears, and a one-sentence summary of how they apply it.

Decision rules:

What specific rules or heuristics does this person use to make decisions? When they say "I always do X" or "I never do Y," catalog those rules.

Vocabulary patterns:

What unique phrases, terms, or jargon does this person use repeatedly? Define each one in their own words. This is their conceptual vocabulary.

Pass 3: Strategy

Positioning analysis:

What is this person's core positioning? Who is their ideal audience? What transformation do they promise their audience? What do they explicitly NOT promise?

Content structure:

How does this person typically structure their videos or content? What hooks do they use? How do they open, build tension, and close? What's their signature move?

Business model reveal:

Based on what they've said publicly, how does this person actually make money? What are all the revenue streams visible in their content? Which one seems to be the biggest?

Pass 4: Synthesis

Three-video shortlist:

If I could only watch 3 videos from this entire channel to get 80% of this person's most valuable thinking, which 3 would you recommend and why?

Hidden gems:

What's the most surprising, underrated, or non-obvious insight buried in their content that most viewers would probably miss? Look for things they mention once but don't expand on.

One-sentence summary:

In one paragraph, summarize this person's entire worldview, methodology, and value proposition as if you were briefing me before a meeting with them.

That last prompt is the one I save. It's the briefing document I keep.

The meta-moves nobody else is teaching

The four passes plus twelve prompts are the foundation. These next moves are what separates someone using this as a tool from someone using it as a research system.

Move 1: Comparative notebooks

Create one notebook per creator. Then run the same question across all of them.

Example: load three operators in the same niche. Ask each notebook: "What's your strongest opinion about how to grow on this platform?" You now have a side-by-side comparison of three experts. You can see where they agree (probably industry truth), where they disagree (their actual differentiation), and where one of them is uniquely right.

This is comparative analysis at the speed of typing.

Move 2: Contradiction hunting

Within one creator, ask: "Where does this person contradict themselves across their content? Find statements that are inconsistent with each other and show me both."

Every public thinker contradicts themselves. Tracking these contradictions tells you two things: where they're still working out their thinking (useful), and where their public persona diverges from their actual beliefs (more useful).

Move 3: Evolution tracking

If you've loaded a creator's content chronologically, ask: "How has this person's focus or messaging shifted from their earliest videos to their most recent?"

Most creators evolve in patterns. Tracking the evolution tells you what they discovered that worked, what they abandoned that didn't, and where they're heading.

Move 4: The "what would they say" prompt

Once a notebook is loaded, ask: "Based on everything you know from their content, what would this person say about [SITUATION YOU'RE FACING]?"

This is genuinely useful. You can effectively ask a creator for their take on your specific situation without their involvement. The answer is grounded in their actual stated positions, with citations. Not perfect, but better than guessing.

Move 5: The cross-domain stack

Load creators from different domains who you think might have hidden overlap. Investor + neuroscientist. Operator + philosopher. Content creator + martial artist.

Then ask: "What do these two people both seem to believe, despite working in completely different fields?"

The patterns that show up across domains are usually the deepest truths. This move surfaces them.

Bonus move: Mount across notebooks in Gemini

In 2026, Google shipped a way to mount NotebookLM notebooks as data sources directly inside the Gemini app. Click the "+" in Gemini's chat input, pick "NotebookLM," select multiple notebooks, ask questions that span all of them. You're now running comparative queries across every creator you've ever studied, in one prompt.

This breaks the single-notebook silo problem. If you stick with the protocol for a year, this is the move that turns your collection into something genuinely powerful.

The storage layer (where the protocol compounds)

This is the part most people skip. They run the prompts, get answers, close the tab. Six months later they're starting from scratch on the next creator.

The compounding move: every time you finish the four passes on someone, save the synthesis output (the "one-sentence summary" prompt above) into a personal database. Mine lives in Notion. Yours can live anywhere you'll actually look.

Over time, this database becomes a personal cheat sheet of how every thinker you've studied actually thinks. You can pattern-match across them. You can pull up "what does X think about Y" in seconds. You can find contradictions between people you respect.

Six months in, the database is more valuable than any individual notebook.

This is the difference between using AI to do research and building a research system that gets smarter over time.

Three uses from the past month

The collaboration filter. Loaded a creator I was considering partnering with. Ran the worldview pass. Found that their public-facing brand and their actual stated positions were misaligned. Killed the partnership conversation in 20 minutes instead of three coffee meetings.

The niche entry decision. Loaded three operators in a niche I was thinking about entering. Ran the comparative move (same prompts across all three). Found the consensus, the disagreements, and the unfilled gap. The gap was the opportunity.

The health protocol research. Loaded a longevity creator's full library before a specific health decision. Asked the "what would they say about" prompt with my specific bloodwork situation. Got the answer with citations, in one session, instead of watching 40 hours of content.

The pattern is the same every time. Pick someone whose thinking you respect, run the protocol, save the synthesis, make better decisions.

What breaks and how to fix it

The transcription is wrong. YouTube auto-captions can be garbled, especially for technical or non-English-fluent speakers. NotebookLM uses these captions. If the answers look off, check whether the source video has good captions. Skip videos with bad captions.

Hit the 50-source cap mid-import. The extension will tell you when the notebook is full. Either prune lower-priority videos or split across two notebooks. Don't try to merge transcripts manually, the quality drops.

The answers are too generic. NotebookLM defaults to summarized, careful answers. Push it. "Be specific. Quote directly. Don't hedge. Show me the citations." Direct prompting gets direct answers.

It cites the wrong video. Happens occasionally with similar-sounding videos. Ask: "Show me the exact quote and which video it's from with timestamp if possible." Forces a precision check.

You can't think of who to study. Start with three people: someone you'd like to be like in 5 years, someone whose business model interests you, someone whose worldview challenges yours. Three different filters. Three different lessons.

When to graduate to the automated version

This protocol is the manual version. It works at scale up to about 10 to 15 creators if you're disciplined about the storage layer.

If you find yourself wanting to: automate the import by topic instead of by channel, generate research briefs on demand, produce audio summaries or infographics automatically, or run this as a scheduled workflow, there's an upgraded version called Claude Code + NotebookLM (filed under PB-011). It connects Claude Code directly to NotebookLM through a single install prompt, so the entire pipeline runs from a natural-language command instead of clicks.

Start here. Move to Claude Code + NotebookLM (PB-011) only when you've hit the limits of the manual workflow.

What this changes

For most of history, deeply learning from someone required either a personal relationship with them or spending weeks consuming their work. The cost of getting inside someone's head was high. So most of us only really got inside the heads of a small number of people in our lives.

This protocol drops that cost by orders of magnitude. An afternoon to be inside someone's body of work. A week to pattern-match across multiple thinkers. A year to build a personal database of how every smart person you've studied actually thinks.

The compounding effect is the real story. Twelve months in, you don't just know more. You know how to ask better questions because you have more reference points. You spot contradictions faster because you've seen them before. You find the gaps in markets and conversations because you've mapped the existing positions.

The people who run this protocol consistently are going to compound knowledge faster than the people consuming content linearly. Same way the printing press separated the people who learned to read from the people who didn't.

Pick three people whose thinking you respect. Run the four passes on each. Save the synthesis. Do this once a month for a year. See what you become.