Gemini Notebook (Formerly NotebookLM): The Free Google Tool That Answers From Your Own Documents

The document pile that broke my usual workflow

Last month I tried to prepare a single research summary from a stack of material that had grown out of control. There were three long PDF reports, a set of lecture slides, a handful of web articles I had saved, and a YouTube video I kept meaning to watch. Every time I opened a general chatbot and asked a question about that pile, I got a fluent answer. The problem was that I could not tell which document the answer came from, or whether the chatbot had filled a gap with something it had read somewhere else entirely. I spent almost as much time checking the answer as I would have spent reading the documents myself.

That experience is the reason Google’s Gemini Notebook caught my attention. It does not try to answer every question in the world. It answers questions about documents you give it, and it shows you where each part of the answer came from. If your material does not cover a question, it is supposed to say so rather than inventing a confident reply. For students, researchers, bloggers, and anyone who works with long documents, that narrower promise is more useful than a broader one.

I have spent the past week working through Google’s own documentation for the tool, its current plans page, and several independent guides that tested the free tier in October 2026. I have not tested the paid Google AI plans myself, and I will say so plainly wherever a price or a paid limit appears below. Everything about the free tier below reflects what Google publishes and what independent testers found on free accounts. Prices are in US dollars, as Google lists them, and availability can vary by country, so check the plans page for your region before you pay for anything.

First, the name change you need to know

If you search for this tool, you will find it under two names. Google launched it as NotebookLM. On July 16, 2026, Google renamed it Gemini Notebook. It is the same product, not a replacement. Old NotebookLM links redirect, existing notebooks remain in place, and the new home is notebook.google.com. If a tutorial still calls it NotebookLM, it is describing the same tool, although its screenshots and prices may be out of date.

The rename matters for a practical reason. Articles written before July 2026 describe older limits, older prices, and a smaller feature set. If you follow one of those guides and find that a button has moved or a limit has changed, that is probably why. Start from the current name, Gemini Notebook, and treat pre-rename pricing as history.

What Gemini Notebook actually does

Gemini Notebook is a research workspace built around sources. You create a notebook for one project, one course, or one question. You add sources to that notebook. Then you ask questions in a chat panel, and the answers are drawn from those sources only, with citations you can click to see the exact passage behind a claim. You can also turn the same sources into other formats: a podcast-style audio discussion, a video overview, a mind map, flashcards, a quiz, a report, or a slide deck.

The three panels on screen explain most of the product. On the left, the sources panel holds everything you uploaded or linked. In the middle, the chat panel is where you ask questions and read cited answers. On the right, the Studio panel is where you generate outputs from your sources. Notes you save sit alongside those panels, so a useful answer or a generated summary can be kept for later instead of disappearing when you close the tab.

That source-first design is the whole point. A general chatbot draws on a vast amount of training data and, in most cases, on web search. Gemini Notebook narrows the field on purpose. It can only use what you gave it, which makes its answers easier to check and, in my view, easier to trust for work that has to be accurate. It also means the quality of the output depends directly on the quality of your sources. Put in a thin or one-sided set of documents, and you will get a thin or one-sided answer, however polished it sounds.

Setting up your first notebook in ten minutes

You need a Google account. There is no separate Gemini Notebook subscription for the free tier, no card is required to start, and the tool runs in a browser at notebook.google.com. A mobile app exists in some regions, but the browser version is the full experience, and that is what I would start with.

1. Create one notebook for one project. Give it a name tied to the job, such as “Biology revision” or “Client onboarding research”. Notebooks cannot see each other, so mixing two projects in one notebook will muddy your answers. Keep them separate from the start.
2. Add your sources. You can upload PDFs, Word documents, text files, Markdown, spreadsheets, presentations, and audio files such as MP3 and WAV. You can also connect Google Docs, Slides, and Sheets from Drive, paste a web link, add a public YouTube URL, or paste text directly. Each source on the free tier can hold up to 500,000 words or 200 MB per uploaded file.
3. Wait for processing to finish. The tool scans and indexes each source before you can question it properly. Large PDFs take longer. If an answer looks thin, check first whether every source has finished processing.
4. Ask a specific question. Vague questions produce vague summaries. Ask for a comparison, a definition, a list of dates, or an explanation of one concept. Then click the citations. A citation that does not support the claim is a warning sign, and checking it takes seconds.
5. Save what matters. Pin useful answers and generated outputs as notes. A notebook that only holds sources is a filing cabinet. A notebook with saved notes becomes a working document you can return to.

One setting is worth changing early. In the notebook menu, Configure Chat lets you choose how the chat behaves. The default mode answers directly. Learning Guide mode is built for studying: it asks guiding questions and encourages you to work through material rather than copying an answer. If you are revising for an exam, Learning Guide is the better choice. If you need a fast factual answer from a report, the default mode is faster.

The Studio panel: where it earns its keep

Chat is useful, but the Studio panel is where Gemini Notebook separates itself from a plain document search. Each output is generated from the same sources, so a mind map, a quiz, and an audio overview should all agree with each other, because they are all grounded in the same material.

Audio Overviews turn your sources into a podcast-style discussion between two AI hosts. There are formats for a deep dive, a short brief, a critique, and a debate, so you can choose between a thorough walkthrough and a quick refresher. This is the feature most people try first, and it is genuinely good for revision on a walk or a commute. It is also the most expensive output in terms of usage, which matters on the free tier, as I explain below.

Video Overviews give you a visual summary of the same material, useful when you need to present a topic or grasp its structure quickly. The fully animated Cinematic Video Overview is the exception: it requires a Pro or Ultra plan, and it is not included on the free or Plus tiers.

Mind Maps draw a branching diagram of the topics in your sources. Clicking a node creates a chat prompt about that concept in context, which is a neat way to move from a broad overview into a specific question without retyping it. For a large subject, a mind map is often the fastest way to see what you have missed.

Flashcards and quizzes are the quiet workhorses. You choose the number of items, the difficulty, and in the quiz case the question types, which include multiple choice, multiple select, fill in the blank, and short answer. Flashcards can be marked as got it or missed it, and you can practise only the ones you missed, ask for an explanation of a card, shuffle the deck, or download it as a CSV file. For revision, these outputs are also cheap on the free usage meter compared with audio and video.

Reports, study guides, and slide decks turn sources into structured documents. Report templates include a briefing document that pulls out key themes, a study guide with a short-answer quiz, essay prompts, and a glossary, and a blog-post style summary. Slide decks and infographics turn the same material into something presentable, although you should still check every slide against its source before you show it to anyone else.

Deep Research is for longer investigations. Instead of answering from a single set of uploaded sources, it can search for additional material and build a fuller research base, which you can then question in the usual way. It is a research mode, not a shortcut: the results still need the same citation checks as any other output.

What it costs in 2026

Gemini Notebook itself is not sold on its own. The free tier comes with a Google account, and paid access arrives through a Google AI plan, a qualifying Google Workspace plan, or a Google Cloud Enterprise licence. That bundling confuses people, so here is the consumer picture as Google lists it for US accounts.

Two points in that table deserve emphasis. First, every tier runs on the same Gemini 3 model and the same per-source ceiling of 500,000 words or 200 MB. Paying more buys room and usage, not a smarter assistant, plus paid-only extras such as Cinematic Video Overviews and advanced sharing. Second, the free tier is genuinely usable. One hundred notebooks with 50 sources each covers most students and most solo researchers. The constraint is volume, not missing features: audio and video outputs consume far more of the usage budget than chat or flashcards, and independent testers in October 2026 found a single audio overview could take a large share of a five-hour usage window, while a set of flashcards took only a few percent.

That usage model has changed over time, and older articles will mislead you here. Google no longer counts usage simply in chats or overviews per day on personal accounts. Its help pages describe compute-based limits: a long chat, a heavy feature, or a complex prompt costs more, and you receive a fresh allowance every five hours up to a weekly cap. Older figures such as 50 chats per day and 3 audio overviews per day come from an earlier system. Treat them as outdated, and watch the meter shown in the product itself before you generate a large output.

Where it beats a general chatbot, and where it does not

The comparison explains why I would not replace a general chatbot with Gemini Notebook. I would use both, for different jobs. When I need to understand a set of documents I already have, Gemini Notebook is the better tool. When I need ideas, a draft, or an answer about something I have not collected sources for, a general chatbot is the better choice. My earlier comparison of Claude Duet, ChatGPT Dots, and Meta Muse made a similar point about proactive agents: the right question is rarely which tool is best in general, but which tool fits the specific job in front of you.

There is also a difference in how each tool handles uncertainty. A general chatbot will usually give you an answer even when it is unsure, and it will often sound confident while doing so. Gemini Notebook is designed to say when your sources do not cover a question. That behaviour is less impressive in a demo and far more useful in real work, because a missing answer is a signal to find a better source, while a confident wrong answer is a trap.

Five ways I would actually use it

1. Revising for an exam. Put one course in one notebook: lecture slides, readings, and your own notes. Start with a mind map to see the structure, use Learning Guide mode to work through weak topics, then drill with flashcards and a quiz on hard difficulty. Save audio overviews for revision on the move, because they cost the most usage.

2. Reading a long report for work. Upload the report and any background documents, then ask for a briefing: key findings, decisions needed, deadlines, and risks. Check each cited passage before you forward the summary. The time saved is real, but only if you keep the checking step.

3. Preparing a blog post or article. I collect source material first, then use Gemini Notebook to compare what two sources say, pull out exact figures, and test whether my draft fairly reflects the material. It does not write the article for me, and I would not want it to: my writing has to reflect my own judgement, not a summary. My guide to Gemini 4 Argon is an example of the kind of model news where checking price and access against primary sources matters more than speed.

4. Onboarding into a new topic. When a client or a project hands you a pile of background reading, a briefing document plus a mind map gives you a map of the territory in an afternoon. Follow it with targeted questions about the parts that affect your decisions, and ignore the rest until you need it.

5. Learning a skill from documentation. For a programming language, a tool, or a certification syllabus, load the official docs and a good tutorial, then use quizzes to test yourself as you go. Gemini Notebook is weaker here than for reading comprehension, because code examples need testing in a real environment, not just summarising. Use it to understand the docs, then practise in the actual tool.

The limits you should know before you upload anything

The first limit is privacy. Your files are processed on Google’s servers, and there is no offline mode. Do not upload confidential client material, unpublished research you cannot afford to expose, medical records, or anything with passwords and private keys. For ordinary study material and public documents, that trade is reasonable. For sensitive work, it is a reason to choose a different tool, or to keep that project out of Gemini Notebook entirely.

The second limit is scope. A notebook can only answer from its own sources, and notebooks cannot query each other. If your answer needs material from two projects, you must combine the sources into one notebook or ask the question twice and compare the results yourself. That separation is a feature for accuracy, but it is a real inconvenience for broad research.

The third limit is accuracy in a subtler form. Source-grounded answers reduce invention, but they do not remove it. A summary can still misread a passage, overstate a claim, or merge two sources that disagree. Citations help only if you click them. If you copy an answer without checking at least the important claims, you have given up the main advantage of the tool.

The fourth limit is the usage meter. On the free tier, a heavy session of audio and video outputs can hit the five-hour allowance sooner than you expect. Plan around it: do your chat and flashcard work first, generate audio last, and save large outputs for when you actually need them. Paying for a plan solves the volume problem, but it does not change the model or the per-source ceiling, so do not upgrade expecting smarter answers.

The fifth limit is availability and change. Google has renamed this product, restructured its plans at Google I/O in May 2026, and changed how usage is counted. Features, prices, and limits described here reflect what was published in October 2026, and they will change again. Before you build a course or a business workflow around Gemini Notebook, confirm the current limits on Google’s own plans page. The broader AI market is moving fast enough that even large companies are repricing their AI bets, as I covered when Alphabet lost $500 billion in a day, and a tool owned by Google is not immune to that pressure.

Who should pay, and who should stay free

Stay on the free tier if you are a student with one or two courses at a time, a solo researcher with a handful of projects, or someone trying the tool for the first time. The free limits are generous enough to learn whether the workflow suits you, and the only cost is the time you spend setting up good sources.

Consider Google AI Plus at $4.99 a month if you regularly hit the source limit, run many notebooks, or want early access to new features without committing to a larger plan. It doubles your room and your usage budget, which is a fair step up for a heavy student or a freelancer.

Consider Google AI Pro at $19.99 a month if you generate audio and video overviews often, manage large research projects, or need Cinematic Video Overviews and advanced sharing. Note that Pro also includes much more than Gemini Notebook, including higher access to Gemini models elsewhere, so judge the price on the whole bundle, not on this tool alone. Students should also check Google’s student offer for AI Pro before paying, where eligible.

Ultra, at $99.99 or $199.99 a month, is for teams and heavy professional use, not for trying out a research tool. If you are reading this article to decide whether Gemini Notebook is useful, Ultra is not your starting point. Start free, and let your usage meter tell you whether you need more room.

My honest recommendation

I would recommend Gemini Notebook to anyone who regularly works with long documents and wants answers they can check. Students revising from course material, researchers synthesising papers, and professionals summarising reports will get the most from it, and the free tier is strong enough that there is no reason not to try it with one real project first.

I would not recommend it as a general writing tool, a replacement for a chatbot, or a place for confidential material. If your main need is drafting, brainstorming, or open-ended questions, a general chatbot serves you better. If your material is sensitive, keep it out of a cloud-processed tool. And if you already pay for a research tool that gives you cited answers from your own library, Gemini Notebook may duplicate what you have rather than add to it.

What I would do, and what I suggest you do, is simple. Pick one project you are working on right now, not a test project. Put its real sources in one notebook, ask five questions you actually need answered, and check every citation. Then generate one mind map and one set of flashcards, and see whether they save you time in the next study or work session. If they do, keep the notebook and build the habit of saving notes as you go. If they do not, you have lost an hour, not a subscription, and you will know exactly why. That small, honest test tells you more about a tool than any feature list, including this one.