How to Use NotebookLM: Turn Your Own Documents Into Answers You Can Check

Most AI chatbots answer from what they learned in training, which is fine for general questions and risky for anything specific to you. NotebookLM works the other way round. You give it your own material — lecture notes, reports, contracts, transcripts, a stack of PDFs — and it answers from those sources, with numbered citations that jump to the passage it used. In July 2026 Google renamed it Gemini Notebook, but most people still search for the old name, and the product is the same. This guide covers how to use NotebookLM properly: which sources to add, how to ask questions that actually use them, when the Audio Overviews are worth your time and where it falls short.

What NotebookLM is (and why it's now Gemini Notebook)

NotebookLM started life at Google I/O 2023 as Project Tailwind, an experiment in helping people learn from their own documents. It grew into Google's research notebook, and on 16 July 2026 Google renamed it Gemini Notebook to tie it more closely to the rest of the Gemini family. Old links redirect, existing notebooks stayed where they were, and the app now lives at notebook.google. You can also open your notebooks from inside the Gemini app, and changes sync both ways.

What makes it different from a general chat assistant:

• Grounded in your sources — answers are built from the documents in the notebook, not from whatever the model half-remembers about the topic.
• Citations you can click — each claim links to the exact passage, so checking takes seconds rather than minutes.
• One notebook per project — sources, chat and saved notes live together, so a dissertation, a client project and a house purchase don't bleed into each other.
• Ready-made outputs — the Studio panel turns your sources into audio and video overviews, study guides, flashcards, quizzes, mind maps, reports and slide decks.

If you mainly want Gemini as an all-round assistant, our guide on how to use Gemini is the better starting point. NotebookLM is for when the answer has to come from a specific set of documents.

Setting up your first notebook

You need a Google account, and you may be asked to verify your age before you can start. Then:

1. Create a notebook and give it a name that describes the project, not the files ("Q3 supplier review", not "PDFs").
2. Add sources — upload files, pick documents from Google Drive, paste a link or paste text directly.
3. Let it process — NotebookLM reads each source and usually offers a short summary and a few suggested questions.
4. Start asking in the chat panel, and save the answers worth keeping as notes.

It accepts a wide range of formats: PDFs, Google Docs, Slides and Sheets, Word and PowerPoint files, plain text and Markdown, CSV, ePub, web pages, public YouTube videos with captions, audio files and images. On the free tier you get up to 100 notebooks with up to 50 sources each, and a single source can hold up to 500,000 words, or 200MB as an upload. Paid Google AI plans raise the number of notebooks and sources per notebook. Google adjusts these limits from time to time, so check the help centre before you plan a huge project around them.

Choosing sources that give good answers

NotebookLM is only as good as what you put in it. Ten relevant, clean sources beat fifty loosely related ones, because every extra document is another place the model can pull a half-relevant sentence from.

• Keep one question per notebook — a notebook for "my thesis on urban heat islands" works; a notebook for "everything about climate" doesn't.
• Prefer originals — the study, the policy document or the meeting transcript, rather than someone's summary of it.
• Clean up scans — badly scanned PDFs produce garbled text, and garbled text produces confident nonsense.
• Label versions — if you upload "Contract v2" and "Contract v4", say so in the source names, or you'll get answers that quietly mix the two.
• Use the source checkboxes — tick and untick sources so a question only draws on the ones you choose.

When you don't have sources yet, the built-in research options can find them. Fast Research searches the web or your Drive for material on a topic; Deep Research browses more widely, writes a report and lets you import the pages it used. Treat either as a reading list, not a verdict: open what it imported and delete the weak pages before you rely on them. For a second, independent pass over the open web, our guide on how to use Perplexity covers searching with citations.

Asking questions that actually use your sources

The chat panel is where most of the value is, and the difference between a vague answer and a useful one is usually the question. Name the sources you care about, ask for the format you want and tell it to say when something isn't covered:

Comparing documents is where NotebookLM beats reading everything yourself:

You can also change how the chat behaves. Under Configure chat there's the default style, a Learning Guide mode that asks you questions and breaks problems into steps instead of handing over answers, and a Custom option where you describe a role and goal yourself, plus a setting for response length. A custom instruction might be: "Act as a sceptical reviewer on a grant panel. Point out weak evidence before strengths." The habits behind a good Gemini prompt apply here too: a clear role, a clear task and a defined output.

Then click the citations. Each number opens the passage it came from, which is the fastest way to catch an answer that sounds right but stretches what the document actually says.

Audio Overviews, Video Overviews and the Studio panel

The Studio panel is the feature that made NotebookLM famous: turn a pile of documents into a podcast-style conversation you can listen to on the bus. It's genuinely useful for a first pass through dense material, and the outputs now go well beyond audio:

• Audio Overviews come in four formats: Deep Dive (the default, two hosts connecting ideas across your sources), Brief (one speaker, the key points in a couple of minutes), Critique (two hosts giving feedback on, say, an essay or a design doc) and Debate (two hosts arguing different sides). You can choose a length, pick from more than 80 languages and add your own instructions.
• Video Overviews turn the same material into a narrated explainer with slides and visuals.
• Mind maps show how the topics in your sources connect, which helps when a subject is new to you.
• Reports, study guides, flashcards and quizzes are built for revision and briefings.
• Infographics, slide decks and data tables give you a first draft of something visual or structured.

The customisation box is what makes an Audio Overview worth listening to. Without instructions you get a pleasant, general chat. With them you get something aimed at you:

Remember that the hosts are a format, not experts. They can flatten nuance, add an upbeat framing the documents don't have, or get a detail wrong. Use the overview to find your bearings, then go back to the sources for anything you'll quote, submit or act on.

Study and work workflows with example prompts

Here's how NotebookLM earns its place day to day, with prompts you can copy and adapt.

Revising for an exam. Upload the lecture slides, your notes and the key readings, then switch on Learning Guide or try:

Our AI prompts for students have more ideas for using AI to learn rather than to do the work for you.

Literature review. Add the papers themselves, not just the abstracts:

Project and meeting notes. Upload a few weeks of transcripts or minutes:

Starting a new job or client. Load the handbook, product documentation and past reports, then ask for "the 15 terms I'll hear in my first week, explained in plain English, with where each one is defined". Share the notebook and the next new starter gets the same searchable briefing.

Spreadsheets and CSV files work as sources too, and Google has begun adding code execution on paid plans so a notebook can run calculations on your data. For serious number work, our AI prompts for data analysis go further.

NotebookLM vs Gemini, ChatGPT and Perplexity

These tools overlap, but each has its home ground:

• NotebookLM / Gemini Notebook — when the answer must come from documents you supply, and you want to check every claim against them.
• Gemini, ChatGPT or Claude — for open-ended writing, brainstorming, coding and general questions where your own documents aren't the point.
• Perplexity — for searching the open web for current information, with citations to the pages it found.

A workflow that plays to each tool's strengths: find sources with a search tool, load the good ones into a notebook, question them there, then take your checked notes into a chat assistant to draft the final email, essay or report. Ready-made structures for that last step are in our prompt templates library.

Limits worth knowing

Knowing how to use NotebookLM well also means knowing where it breaks:

• It can still be wrong — grounding reduces errors but doesn't remove them. It can misread a table, merge two sources or overstate a cautious finding. Our explainer on AI hallucinations covers why.
• Your sources set the ceiling — if the documents are outdated, biased or incomplete, the answers will be too, delivered just as confidently.
• It isn't a search engine — outside the research features, it knows nothing about what isn't in your notebook.
• Overviews simplify — Audio and Video Overviews are summaries in a chatty format, not a substitute for reading the material.
• Limits change — source counts, daily generations and which features need a paid plan shift regularly.
• Privacy is your call — read Google's terms for your account type and check your organisation's policy before uploading client files, health records or anything confidential. Work and school accounts may have settings controlled by an administrator.

Used within those limits, NotebookLM is one of the most practical AI tools around: a research assistant that has read exactly what you gave it, shows its working and lets you check every line. Pick a focused set of sources, ask specific questions, click the citations and keep the notes that hold up.

References

• Google — NotebookLM is now Gemini Notebook
• Gemini Notebook Help — Add or discover new sources for your notebook
• Wikipedia — Gemini Notebook