How to Turn YouTube Research Into Reports With Claude Code + Gemini Notebook

I tell my students the same thing every semester. The tool doesn't matter. The pipeline does.

Most people using AI right now have half a pipeline. They ask a question, get an answer, and stop. No sources. Nothing they could hand to someone else and say "this is grounded in something real, go check it yourself." That's fine for a quick question. It falls apart the moment you're actually trying to research something.

This guide fixes that for one specific case: researching a topic across YouTube without watching 20 videos yourself. By the end, Claude Code scrapes the top videos on any topic, hands them to Gemini Notebook, and pulls back a report, a slide deck, or an infographic from a single prompt. No copy-pasting transcripts. No ten browser tabs open at once.

You don't need to know how to code. You need to be willing to type a few commands into a terminal, follow along once, and pay attention the first time something looks wrong. After that, it's one prompt.

Three things, all free.

No paid subscriptions required for any of this.

Claude Code can't pull YouTube data on its own. It needs one tool: a command-line program called yt-dlp. It reads public video metadata, titles, view counts, upload dates, without needing YouTube's official API or a login of its own.

Open your terminal inside Claude Code and run:

Once that finishes, tell Claude Code what you actually want researched:

Claude Code writes and runs the scraping command itself. You're not writing Python. You're describing what you want and watching it happen.

This step matters more than it looks like it should.

Claude Code hands you back a table: titles, channels, views, links. Read it. Does it actually match what you asked for? Wrong niche, outdated videos, an obvious spam channel padding the list, catch it here, not three steps from now.

If something's off, just say so. "Drop anything shorter than 5 minutes." "That third channel isn't relevant, exclude it." Claude Code re-runs the search with your correction. Everything after this step builds on these video links being the right ones, so this is the one place worth slowing down.

Gemini Notebook is Google's free research assistant, renamed from NotebookLM in July 2026. Same tool, same features, just a new name folding it into the Gemini brand. Upload sources, and it answers questions grounded in exactly what you gave it. No guessing, no filling in gaps with something it half-remembers.

Here's the catch. Gemini Notebook doesn't have an official way for other tools to talk to it. Someone solved that anyway. A developer named Teng Lin built an open-source package called notebooklm-py that lets Claude Code control Gemini Notebook directly, no public API needed.

Install it the same way as before:

Then log in, once:

A browser window opens. Sign in with your Google account like you normally would. Claude Code stays connected after that. You won't be asked again unless you sign out yourself.

Tell Claude Code to take the video links from Step 1 and load them straight into a new notebook in Gemini Notebook.

Gemini Notebook holds up to 50 sources per notebook, so this scales well past 10 videos if your research needs more depth. And because the sources go in through the command line, you never touch Gemini Notebook's upload button once. That's the whole point of wiring these two together instead of using them side by side.

This is where the pipeline pays off. Gemini Notebook has already read every transcript you loaded. Now you ask Claude Code to pull insight out of that and turn it into something you can actually use.

Gemini Notebook does the actual thinking. It reads through every transcript and finds the pattern across all ten. Claude Code just asks the question and formats what comes back. That division matters more than it sounds. Your token usage on Claude Code's side stays close to nothing, because the heavy analysis happens inside Google's system, not yours. You're not paying twice for the same work.

Before this pipeline existed, doing the same research meant opening ten videos, either watching them or hunting for transcripts, pasting chunks into a chat window one at a time, and hoping you remembered which claim came from which video. That's an hour, easily, before you've written a single sentence of your own analysis.

Now it's five minutes of setup and one prompt. The quality difference is real too. Because Gemini Notebook answers are grounded in the actual sources, you get citations back, not a confident paragraph you have to fact-check yourself later.

A few things trip people up the first time through. None of them are hard to fix once you know what's happening.

None of these mean something's broken. They mean the setup step needs a retry, which takes seconds once you know what to look for.

Once you've run Steps 1 through 5 manually and confirmed each piece works, compress the whole thing into a single request:

Five minutes of setup, once. After that, one prompt. Claude Code handles scraping, loading, and formatting. Gemini Notebook handles the actual reading and reasoning.

I use a version of this before deciding whether a new AI tool release is worth writing a full guide about. If ten different creators are all pointing at the same feature, that's a signal worth trusting. If they're all over the place, that tells me something too, usually that the tool isn't as settled as the hype suggests.

This isn't a shortcut around doing the reading. It's a way to do the reading across ten sources instead of one, without losing your whole afternoon to open tabs.

That's all I'm asking.

Become a paying subscriber to get access to this post and all other premium content.