# NotebookLM > Turn a pile of PDFs, transcripts, and meeting recordings into a grounded knowledge base you can chat with, map, and listen to _20 min · beginner · track: chat-agent · id: notebooklm_ > **Team:** > > You have a folder somewhere with the things that actually matter for your > work right now: the contract, the strategy doc, three weeks of meeting > notes, two transcripts, the slides from the customer call. Too much to read > end to end, too important to skim. NotebookLM is the tool for exactly that > pile. You drop the files into a "notebook", and from then on you ask questions and the answers come back grounded in those sources, with clickable citations to the exact paragraph. It does not roam the internet. It does not invent. It cites the line. ## Grounding: chat with your data, not the web **NotebookLM is the inverse of a chatbot: a workspace whose entire world is the sources you give it.** Where [Gemini](/course/gemini) reaches across the whole internet plus your Workspace, NotebookLM is restricted to the files in the notebook. Under the hood it is RAG (retrieval-augmented generation): the model finds the relevant chunks, builds the answer from them, and shows you which chunks it used. Two things follow from that. Hallucinations drop, because the model is restricted to material you gave it. And the answer is auditable: every claim has a `[1]` next to it that jumps to the source passage. You can verify in two seconds whether the model read the doc the way you would have. The sweet spot is a tightly scoped notebook for a specific question or project, not one mega-notebook with everything you have ever read. Five to twenty closely related sources beats fifty loosely related ones, every time. When the topic shifts, build a new notebook. > **Tip:** > > **Try it.** Go to [notebooklm.google.com](https://notebooklm.google.com), create a notebook for a topic you are working on right now, and drop in 3 to 5 sources. Ask a cross-document question you would normally need to read all of them to answer ("what are the open questions across these, and which source raised each one?"). ## Multimodal sources, including the meeting recording **A notebook accepts almost any readable or listenable source.** Not just PDFs and Docs: - Google Docs, PDFs, plain text, slide decks, web URLs. - Audio files: MP3s of meetings, interviews, lectures. NotebookLM transcribes and indexes them, so you can ask "what did the customer say about pricing in the recording from last Tuesday". - Images and visuals inside slides, parsed for both their text and their content. The free tier holds up to 50 sources per notebook; NotebookLM Plus (bundled with Google AI Pro / Ultra) raises that to 300. Both are plenty for almost any real project. There is also a "Discover sources" button that lets the notebook search the web for related material and add it for you, useful when you are missing a public reference and do not want to leave the tool. > **Tip:** > > **Try it.** Find an audio recording in your work (a meeting recap, a customer interview, a podcast episode) and drop it into a notebook alongside two related docs. Ask: "What is the most important thing the speaker said, and where in the recording did they say it?" ## The Studio panel: pick the output that fits the moment **The same notebook can produce several different artifacts from the Studio panel on the right.** Same sources, different shapes: - **Audio Overview**: a two-host podcast version of your sources. Banter, examples, pacing. The original viral feature, and still the right format for a commute or a walk. Interactive mode lets you interrupt the hosts and ask follow-ups like calling into a radio show. - **Video Overview**: a narrated slide-style explainer with visuals pulled from your sources. Good for sharing context with someone who will not read the docs but will watch a three-minute clip. - **Mind Map**: a clickable, expandable map of the topics across your sources. Useful when you do not yet know what question to ask, you are still figuring out the shape of the material. - **Briefing doc, Study guide, Timeline, FAQ**: structured text outputs the model writes from your sources. Pick the one that matches what you actually need to send. Two ways to make any of these more useful. Give a focus prompt before generating ("focus on the financial risks", "explain it as if to a new hire", "be skeptical of the claims") and the output steers accordingly. And save useful chat answers as notes inside the notebook so they become first-class material the next output can draw on. > **Tip:** > > **Try it.** Generate an Audio Overview from your notebook. Before clicking generate, type a focus prompt: "Focus on the unresolved questions and the decisions still to be made. Treat me as someone who has read the sources but lost the thread." Listen to the first three minutes on your next walk. ## Notebooks inside Gemini and Gems **A notebook can be plugged into a Gemini chat or Gem to ground the conversation in the same sources.** The notebook becomes a reusable asset rather than a one-shot. A "Product Expert" Gem backed by your product docs answers customer questions in your team's voice, citing your actual material. A "Deal Brief" Gem backed by a notebook of contracts and meeting notes briefs you on a customer in two minutes. > **Tip:** > > **Try it.** Build a notebook for a topic you answer questions about often (a customer, a product area, your team's playbook). Attach the notebook to a fresh Gem with a one-line persona ("you are a product expert; cite the docs in every answer"). Ask the Gem the kind of question someone outside the topic might ask, and watch the citations show up. > **Warning:** > > NotebookLM processes data on Google's servers. Treat it the way the [AI-first mindset](/course/ai-first-mindset) data tiers describe: fine for public material and most internal docs, not for customer or production data unless your company's policy explicitly approves it. ## Hands-on 1. Go to [notebooklm.google.com](https://notebooklm.google.com) and create a new notebook. Pick one real topic you have on your plate this week (a project, a customer, a decision). Upload 3 to 8 sources for that topic: meeting notes, a strategy doc, a transcript, two PDFs, the latest deck. Tightly scoped beats comprehensive. 2. Ask a cross-document question that would normally take you twenty minutes to answer: ``` What are the key risks or open questions that come up across these sources? Group by theme. For each, cite which source raised it. ``` Read the answer with the citations panel open. Click two citations to confirm the model is reading the docs the way you would. 3. Open the Studio panel and generate two outputs from the same notebook: an **Audio Overview** and a **Mind Map**. Before clicking generate on the audio, give it a focus prompt: ``` Focus on the unresolved questions and the decisions still to be made. Be direct, skip the polite intro. Treat me as someone who has read the sources but lost the thread. ``` Listen to the first three minutes on a walk. Then come back and click around the mind map. Notice which questions come back to you that you had forgotten you were holding. ## Reflect - Pick the messiest folder in your work right now: the project, the deal, the topic where information is spread across too many places. What is the notebook you would build, and what is the one question you would ask it first? - When have you summarized a doc in a meeting and gotten it slightly wrong? With a notebook open and citations live, that mistake gets a lot harder to make. Where in your week could that habit save you?