How to Use NotebookLM With YouTube Videos

It Reads the Transcript, Not the Visuals

Gemini Notebook (formerly NotebookLM) can add a public YouTube video as a source, but it's important to understand what actually happens: it works from the video's captions or transcript, not from the visual content itself. This means it works excellently for lecture recordings, conference talks, and podcast-style videos where the spoken content carries the information, and much less well for anything where the actual information is visual — a diagram drawn on screen, a demonstration, a chart shown without being read aloud.

Adding one is simple: paste the YouTube URL directly into the Add Source dialog, the same way you'd add a webpage or PDF. The video needs public captions available — either creator-uploaded or YouTube's auto-generated ones — for this to work at all, since there's no transcript to read otherwise.

Why Some Videos Work Great and Others Don't at All

A video with clean, accurate captions — most established educational channels, professionally produced conference talks, and podcasts with a transcript — becomes a fully searchable, citable source in your notebook, exactly like a PDF or webpage. A video without any captions at all can't be added as a meaningful source, since there's no text for the tool to actually work from. Somewhere in between, a video with auto-generated captions full of errors (common with heavy background music, multiple overlapping speakers, or strong accents auto-caption tools struggle with) will produce a source Gemini Notebook can use, but with a real chance of small inaccuracies inherited directly from the caption errors.

This is worth checking directly before relying on a video source for anything important: open the video's own caption track in YouTube first and skim it for accuracy, since a garbled caption track will produce a correspondingly garbled understanding no matter how good the underlying model summarizing it is.

A video's caption quality directly determines how reliable it is as a Gemini Notebook source — worth checking before trusting it.

Adding a YouTube Video and Getting the Most From It

Which Kinds of Videos Are Actually Worth Adding

Lecture recordings, conference talks, interviews, and podcast-style videos are excellent candidates, since the spoken content is the actual point of the video. Tutorial videos demonstrating a visual process — cooking, a software walkthrough, a physical repair — are poor candidates, since the transcript alone ("now I'm going to click here" or "add this ingredient") loses most of the actual information that lived in what was shown on screen rather than said aloud. When in doubt, ask whether you could get the same value from just reading a transcript of the audio with no video at all — if yes, it's a strong candidate; if the video's value is mostly visual, it isn't.

Very Recent Uploads and Live Streams Often Don't Work Yet

A video uploaded very recently may not have auto-generated captions ready yet, since caption generation isn't instantaneous, and a live stream in progress has no complete transcript to work from at all until well after it ends. If a recent upload fails to add properly as a source, waiting a few hours for YouTube's caption generation to complete before trying again is usually the actual fix, rather than assuming the video is fundamentally incompatible.

Can Gemini Notebook Watch and Understand What's Shown in a Video?

No — it works from the transcript or captions only, not from analyzing the visual frames of the video itself. This is the single most important thing to understand before relying on a video source: if the actual information you need is something shown on screen rather than said aloud, Gemini Notebook's video source won't capture it, and you'll get an incomplete or misleading picture without any obvious warning that something visual was missed.

Citations Link Back to a Timestamp, Not Just the Video

When Gemini Notebook cites a claim from a video source, the citation typically links back to the approximate point in the video where that content appears, not just to the video as a whole — which means you can jump directly to the relevant moment to verify a claim in its original spoken context, the same courtesy the tool extends to a specific page in a PDF source. This is worth using actively rather than trusting a citation blindly, especially for anything where the exact wording or framing in the original video matters.

Start With One Video You Already Know Well

Before relying on this feature for something you don't already understand, add one video you're already familiar with as a test source, ask it several questions you know the answers to, and check both the accuracy of the answers and whether the citations correctly point back to the right part of the video. This quick trial run builds a concrete sense of how reliable this specific feature is before you lean on it for material you're trying to learn for the first time.

Copyright-Restricted or Private Videos Won't Work at All

Beyond the caption-quality issue, a video that's unlisted with restricted access, region-locked, or flagged with certain copyright claims may fail to add as a source even with perfect captions, since Gemini Notebook needs to actually access the public video page to pull its transcript. If a video from a legitimate educational channel unexpectedly fails to add, checking whether it's fully public (not unlisted or restricted in your region) is worth doing before assuming the feature itself is broken.

A Playlist Can Be Added as Multiple Individual Sources

Gemini Notebook doesn't currently accept an entire YouTube playlist as a single source in one action, so for a multi-part lecture series or a course taught across several videos, each video needs to be added individually as its own source. This is more setup work upfront for a long series, but it also means each video gets its own distinct citation, letting you trace an answer back to exactly which part of the series it came from rather than a vaguer reference to "the playlist" as a whole.

For a series with a genuinely large number of parts, it's worth prioritizing which specific videos to add first based on which lectures actually cover exam-relevant material, rather than mechanically adding every video in order, since the setup time scales directly with how many individual videos you choose to include.

Comment Tracks and Live Chat Aren't Part of the Transcript

If a video's actual value includes viewer questions answered live, comments discussed on screen, or an on-screen chat during a livestream replay, none of that makes it into the caption transcript Gemini Notebook reads — it only captures the speaker's own words. For a recorded Q&A session where the questions themselves matter as much as the answers, this can leave a meaningful gap in what the notebook actually understands about the video's full content.