--- title: "What Is NotebookLM? A 2026 Review of Google's AI Notebook" description: "NotebookLM, now Gemini Notebook, answers from your own sources. What it does well, where it falls short, and how to feed it bulk YouTube transcripts." canonical: "https://transcript.im/blog/notebooklm" markdown: "https://transcript.im/blog/notebooklm.md" author: "Leo" category: "NotebookLM" datePublished: "2026-10-20T17:19:57.989Z" dateModified: "2026-10-23T05:28:10.676Z" --- NotebookLM is Google's AI notebook for working with your own material. You give it a set of sources, such as PDFs, Google Docs, web pages, audio files, or YouTube videos, and it answers questions about that set, with a citation on each claim that points back to the passage it came from. It also turns the same sources into study guides, flashcards, quizzes, mind maps, and spoken summaries. On July 16, 2026, Google renamed it Gemini Notebook. The product, the notebooks, and the shared links stayed the same, which is why most people still search for it as NotebookLM. This review covers what NotebookLM actually does, where it is strong, where it stops, and what the limits look like on the free tier. It also looks closely at one job a lot of people bring to it: learning from YouTube. NotebookLM can read a YouTube video, but only through the video's captions, one link at a time. That is the point where a [YouTube transcript](/) tool like transcript.im fits in, and the second half of this article compares the two honestly: what each one is for, and how to use them together when you have a whole playlist or channel to get through. ## What Is NotebookLM? NotebookLM started as Project Tailwind at Google I/O 2023. Google describes it as a research and learning tool grounded in your sources, and by mid-2026 it reported more than 30 million people and 600,000 organizations using it. The idea that set it apart from a general chatbot was simple: the model answers from the documents you put in the notebook, not from whatever it remembers from the open web, and it shows its work with numbered citations. ![NotebookLM is now Gemini Notebook: the same notebooks, sources, and citations under a new name](/images/blog/notebooklm/gemini-notebook.jpg) The rename to Gemini Notebook did not create a new product. Google's [announcement](https://blog.google/innovation-and-ai/products/gemini-notebook/notebooklm-gemini-notebook/) calls it "the same standalone product," now more closely tied to the Gemini app, where notebooks can also be created and synced. Existing notebooks and shared links redirect to the new name. The same announcement added a secure cloud computer to each notebook, so it can write and run code for data analysis grounded in your sources. That feature reached Google AI Ultra subscribers first and is rolling out to Pro users. If you read "Gemini Notebook" in a recent guide and "NotebookLM" in an older one, they describe the same tool. This article uses NotebookLM because that is still the name most readers type. A few terms recur throughout: - **Notebook**: a project container. Each notebook holds its own sources, chat, and generated outputs. - **Source**: one item you add, such as a PDF, a Google Doc, a web page, an audio file, a pasted block of text, or a YouTube video. - **Chat**: questions and answers grounded in the notebook's sources, with citations. - **Studio**: the panel that turns sources into formats, including Audio Overviews, Video Overviews, mind maps, reports, study guides, flashcards, quizzes, slide decks, and infographics. The "LM" stands for language model. The name was always a description of the idea: a notebook with a language model that reads only what you gave it. ## How NotebookLM Works Using NotebookLM comes down to three moves, and the quality of the result depends mostly on the first one. For a full walkthrough with prompts, see our guide on [how to use NotebookLM](/blog/how-to-use-notebooklm). ![How to use NotebookLM in three steps: add sources, ask questions with citations, and generate study formats in Studio](/images/blog/notebooklm/how-to-use-notebooklm.jpg) ### 1. Add sources Create a notebook and add what you want to work with. Uploads can be PDFs, text and Markdown files, Word documents, and audio files such as MP3 and WAV. From Google Drive you can add Docs, Slides (up to 100 slides), and Sheets (limited to 100,000 tokens). You can paste web URLs, and when you have several, you can paste them together separated by spaces or new lines. You can paste copied text as its own source. YouTube links are accepted too, with conditions covered below. Audio is transcribed when you import it, and the text is what NotebookLM keeps. ### 2. Ask questions The chat answers from the selected sources. Each claim in the answer carries a numbered citation; hovering or clicking it shows the passage it came from. You can tick sources on and off to narrow a question to one document or widen it to the whole notebook. Good questions are specific: "Where do these two papers disagree about sample size?" works better than "summarize everything." ### 3. Generate formats in Studio The same sources can become a study guide, a briefing report, a timeline, flashcards, a quiz, a mind map, a slide deck, an infographic, or an Audio Overview, which is a podcast-style conversation between two AI hosts about your material (our [NotebookLM Audio Overview](/blog/notebooklm-audio-overview) guide covers its formats and options). Video Overviews add narrated slides. These outputs are where NotebookLM feels different from a chat window: one well-built notebook can feed several ways of reviewing the same material. Everything downstream depends on what goes in. A notebook built from three clear sources gives tighter answers than one stuffed with forty loosely related links. The citations only help if the cited sources deserve trust. ## Where NotebookLM Is Strong ### Answers you can check The single best feature is the citation. When an answer says a lecture defined a term a certain way, you can click through to the exact line and confirm it. That turns NotebookLM from a confident-sounding assistant into something closer to a research aid. You still have to read the citation, but the checking takes seconds instead of a search through a 90-page PDF. ### A bounded set of material NotebookLM is at its best when the scope is clear: one course, one research question, one client's documents, one book. Inside that boundary it can compare arguments, find where a topic appears across ten sources, and pull out definitions, dates, and quotes without wandering into the open web. ### Many formats from one set Students use it for flashcards and quizzes. Researchers use briefing reports and mind maps to see the structure of a field. People who learn by listening use Audio Overviews on a commute. The same notebook serves all of them, and nothing has to be re-uploaded to switch formats. ### Mixed source types A real project rarely comes in one format. NotebookLM takes slides, a recorded talk, a web article, and a PDF of notes in the same notebook and treats them as one body of text. ### Sharing Notebooks can be shared with a class, a study group, or a team, so everyone asks questions against the same sources instead of passing around summaries of summaries. ### A free tier that covers a lot Many students and individual researchers never need to pay. The free limits are generous for one course or one project at a time. ## Where NotebookLM Falls Short ### Grounded is not the same as correct Citations make answers checkable, not right. NotebookLM can skip an exception, flatten a disagreement between two sources, or describe a detail slightly wrong while still pointing at the right paragraph. For anything that matters, open the citation and read the surrounding text. ### It does not judge your sources If a source is outdated, biased, or simply wrong, NotebookLM will answer from it faithfully. The tool has no opinion about which of your sources deserves more weight unless you tell it. ### It only sees text For YouTube in particular, NotebookLM imports the caption text and nothing else. Slides on screen, a diagram on a whiteboard, a code sample shown in the video, or a chart a speaker points at are not part of the source. If the important information was shown rather than said, the notebook does not have it. ### It depends on captions for YouTube A video without captions cannot be added. A video uploaded less than 72 hours earlier may not import yet. The error people run into most is a version of "transcript not available," and there is no option inside NotebookLM to transcribe the audio instead. ### One link, one source There is no import for a whole YouTube playlist or channel. Each video goes in as its own source, which collides quickly with the per-notebook source cap. ### Passive learning risk A polished study guide feels productive. Reading it is not the same as recalling the material or solving a problem. NotebookLM works best as a guide to what to study, not as a replacement for doing the work. ### Limits arrive mid-project Daily chat and Studio usage are capped by plan, and a heavy research day can hit them. The source cap is the one people meet first. ## Is NotebookLM Free? The Limits That Matter Yes. [NotebookLM is free](/blog/is-notebooklm-free) for anyone with a supported Google account, and for one course or one research topic at a time it is usually enough. The limits that shape how you build a notebook are these: ![Is NotebookLM free: the free tier allows 50 sources per notebook, 500,000 words per source, and 200 MB per uploaded file](/images/blog/notebooklm/is-notebooklm-free.jpg) | Limit | Free tier | | --- | --- | | Sources per notebook | Up to 50 | | Size of one source | Up to 500,000 words | | Size of one uploaded file | Up to 200 MB | | Chat and Studio usage | Daily limits that vary by plan | These figures come from Google's [Gemini Notebook help page on adding sources](https://support.google.com/gemininotebook/answer/16215270). Paid Google AI plans and eligible Workspace and Education editions raise the per-notebook source cap and the daily usage limits. Google adjusts plan details from time to time, so check its current plan page before paying for an upgrade. Two limits deserve attention when your material is video. The 50-source cap counts every YouTube link as one source, so a 60-video course does not fit in one free notebook if each video is added separately. The 500,000-word ceiling per source, on the other hand, is large: a one-hour talk produces roughly 8,000 to 10,000 words of transcript, so a single text file can hold the transcripts of dozens of videos and still count as one source. That difference is the basis of the bulk workflow later in this article. On privacy, Google states that notebook content is not used to train its foundation models unless you submit feedback, and that eligible Workspace and Education accounts get additional protections. That does not make every upload appropriate. Follow your school's, employer's, and copyright rules before adding restricted material. ## NotebookLM and YouTube: What Actually Gets Imported Learning from YouTube is one of the most common reasons people open NotebookLM, and it is also where the most confusion starts. Here is what Google's help page says happens when you [add a YouTube video to NotebookLM](/blog/notebooklm-youtube): ![NotebookLM YouTube import: a public video with captions becomes a text source, while a video without captions shows transcript not available](/images/blog/notebooklm/notebooklm-youtube.jpg) - **Only public videos with captions are supported.** Captions can be uploaded by the creator or auto-generated by YouTube. - **Only the text transcript is imported.** The picture, the on-screen text, and the audio itself are not part of the source. - **Recent uploads may fail.** Videos uploaded less than 72 hours earlier may not be available to import. - **Length is not the limit.** A video of any length works unless its caption file passes 500,000 words. - **Imports can fail for other reasons.** Google lists missing caption files, videos flagged as potentially unsafe, and languages that are not supported yet. In practice this means NotebookLM is reading the same caption track you would see in YouTube's own transcript panel. When a creator wrote careful captions, the notebook gets clean text. When the captions are auto-generated, the notebook inherits their mistakes: misheard names, missing punctuation, and technical terms spelled as whatever the speech model guessed. When there are no captions at all, which is common for screen recordings, short uploads, interviews from smaller channels, and many non-English videos, there is nothing for NotebookLM to read. Volume is the other issue. A YouTube link is one source. A course playlist with 40 lessons is 40 separate pastes and 40 of your 50 slots, and if a few of those lessons have no captions, the course arrives with holes in it. NotebookLM does not expand a playlist or channel link into its videos. None of this is a flaw in what NotebookLM is trying to be. It is a notebook for reasoning over text. Getting text out of video is a different job, and it is the job transcript.im does. ## NotebookLM vs transcript.im: Two Different Jobs It is tempting to compare the two as rivals because both touch YouTube and both use AI. They sit at different points of the same workflow. transcript.im turns a video into words. NotebookLM reasons over words you already have. ![YouTube transcript vs NotebookLM: transcript.im turns video links into timestamped text, and NotebookLM answers questions about that text](/images/blog/notebooklm/youtube-transcript.jpg) | | transcript.im | NotebookLM | | --- | --- | --- | | Main job | Get the spoken words out of a video or audio file | Understand and question a set of sources | | Input | A video link from YouTube, TikTok, Instagram, Facebook, X, or LinkedIn, or an uploaded file | Documents, web pages, audio, pasted text, and captioned YouTube videos | | Video without captions | Transcribes the audio with AI | Cannot import it | | Playlists and channels | One link queues every video and returns a labelled set | One video per source, no playlist or channel import | | Output | Timestamped transcript to read, search, copy, translate, or export as TXT, SRT, VTT, Markdown, or JSON | Cited answers, study guides, flashcards, quizzes, mind maps, Audio and Video Overviews | | Timestamps | Kept on every line and in SRT and VTT exports | Not part of an imported YouTube source | | Best when | You need the exact wording of a video, a set of videos, or a file | You need to study, compare, and question material you already have as text | ### Where NotebookLM wins If your material is already text, such as papers, notes, slides, and articles, NotebookLM is the better place to think. Cited answers across many documents, study formats generated from one notebook, Audio Overviews, and sharing with a group are its core strengths, and transcript.im is not built around a multi-source notebook like that. transcript.im does include an AI summary and chat on a transcript, but it is focused on the video you just transcribed, not on fifty mixed documents. ### Where transcript.im wins If the material is video, transcript.im is the step NotebookLM is missing. It reads the creator's caption track when one exists and transcribes the audio when it does not, so a video NotebookLM rejects still becomes text. It keeps timestamps, so a quote can be traced back to the moment it was said. It exports files in the formats other tools expect, translates a transcript while keeping its timestamps, and works on platforms NotebookLM does not read at all, including TikTok, Instagram, and LinkedIn videos. And it takes a playlist or channel link as one job instead of forty. ### Where both fall short Neither tool sees what is only on screen. A transcript records speech, so slides that are never read aloud, code shown without narration, and silent demos are not in either one. For those videos, keep the video open next to your notes. The practical answer is not choosing one. Use transcript.im to get complete, timestamped text out of your videos, then give that text to NotebookLM as sources. ## Bulk YouTube Transcripts for NotebookLM This is the workflow for anyone with more than a handful of videos: a course playlist, a conference channel, a podcast back catalog, or a set of competitor videos you need to analyze. It solves the three problems from earlier, which are videos without captions, one-link-one-source, and the 50-source cap, in one pass. ![Bulk YouTube transcript workflow: one playlist link becomes a labelled set of transcripts, merged into a few NotebookLM sources](/images/blog/notebooklm/bulk-youtube-transcript.jpg) ### 1. Paste the playlist or channel link On the [YouTube playlist transcript](/youtube-playlist-transcript) page, one playlist URL queues every video in the order the playlist sets. On the [YouTube channel transcript](/youtube-channel-transcript) page, one channel URL queues its public uploads. The videos are processed in parallel, so a long course comes back as a set rather than one video at a time. ### 2. Let the gaps fill themselves Each video's caption track is read when it exists. Any video without captions is transcribed from its audio, so a missing caption track no longer leaves a hole in the course. Each transcript is labelled with its video title and keeps its timestamps. ### 3. Check the parts you will rely on Search across the whole set for the terms you care about and spot-check names, numbers, and technical terms against the video. Auto-generated captions and AI transcription both mishear unusual names; fixing them now means NotebookLM cites the right spelling later. ### 4. Export in a format NotebookLM reads well Plain TXT or Markdown works best for a notebook. Markdown keeps each video's title as a heading, which gives NotebookLM clear boundaries between lessons and gives you readable citations. ### 5. Group transcripts to fit the source cap Because one source can hold up to 500,000 words, you do not need one source per video. Combine related transcripts into a few files, for example one file per course module or per month of uploads, and keep each video's title as a heading inside the file. A 60-video course can fit in six or eight sources instead of 60, leaving room in the notebook for slides, readings, and your own notes. ### 6. Add the files to NotebookLM and ask across all of them Upload the files as sources. Now questions like "Which lessons cover pricing?" or "How did the speaker's advice on this change between the first and last episode?" can be answered across the whole set, with citations into the transcript text, and the timestamps in that text lead you back to the exact moment in the original video. The same transcripts are reusable outside NotebookLM. The files work in ChatGPT, Claude, or Gemini, in a notes app, or in a spreadsheet, and SRT or VTT exports can go straight into a video editor. Doing the transcription once and keeping the files means you are not locked into one notebook. A few practical notes: - **Keep the original language for analysis.** Transcribe in the language spoken. If you need another language for a study group, translate the transcript afterwards; the timestamps stay attached to each line. - **Name files by scope.** "Module 3 - Lessons 12-18.md" makes the source list readable and makes it obvious which source a citation came from. - **Refresh, don't rebuild.** For a channel that keeps publishing, run the new uploads and add one new file to the notebook instead of starting again. ## Who Should Use NotebookLM **Use NotebookLM if** you study a course with many readings, compare papers or interviews for research, prepare for exams with flashcards and quizzes, onboard into a new field from a known set of documents, or want to listen to your material as an Audio Overview. It is one of the best free tools available for working inside a defined set of sources. **Pair it with a transcript tool if** your material lives in video. Lectures, conference talks, podcasts published on YouTube, tutorials, interviews, and creator content all need to become text before NotebookLM can reason over them reliably, especially when captions are missing, when the set runs to dozens of videos, or when you need timestamps to find the moment again. **Look elsewhere if** you need guaranteed answers without checking, formal assessment, hands-on practice, or work that depends on what is shown on screen rather than said. NotebookLM also does not go looking for new material on a schedule; it works with what you give it. ## The Verdict NotebookLM is worth using. It is the clearest example of an AI tool that stays inside your material, shows where each answer came from, and turns one set of sources into several ways to learn it. The rename to Gemini Notebook changed the label, not the reasons to use it. Its weak points are predictable: it trusts your sources, it can still miss a detail, the free tier caps a notebook at 50 sources, and for YouTube it sees only caption text, one video at a time. The way to get the most out of it is to give it better input. For documents, that means choosing sources carefully. For video, it means getting complete text first: run the playlist or channel through transcript.im, keep the timestamps, group the transcripts into a few files, and let NotebookLM do what it is best at.