How to Use NotebookLM — Complete Guide (2026)

NotebookLM is Google's AI research tool that answers questions using only the sources you give it, and cites the exact passage behind every claim. To use it: create a notebook, add sources (PDFs, Google Docs, links, YouTube transcripts, audio files), then ask questions, generate study guides, quizzes, mind maps, or turn the material into an Audio Overview podcast. Because every answer is grounded in your own documents rather than the open web, it fabricates far less than a general chatbot does — and it shows you exactly where each answer came from.

This is a step-by-step guide to getting real value out of it, whether you're studying for an exam, doing literature review, or trying to make sense of a pile of contracts.

---

What NotebookLM Is (and Isn't)

NotebookLM is source-grounded AI, sometimes called a "retrieval-augmented" or RAG-style tool (retrieval-augmented generation: the model looks things up in a defined set of documents instead of answering purely from its training data). Unlike ChatGPT or Gemini's general chat mode, which draw on broad training data and, if enabled, live web search, NotebookLM answers only from the documents you give it — and footnotes each claim with a link to the source passage it pulled from.

Think of the difference like the difference between asking a well-read friend a question off the top of their head, versus asking a research assistant who has actually read the specific case file, textbook, or contract in front of you and will only tell you what's in it. The first is faster and more flexible. The second is slower to set up but far more trustworthy when accuracy to your material is what matters — exam prep, contract review, onboarding documentation, or reconciling five conflicting research papers.

Google built NotebookLM at its experimental Labs division and has steadily folded it into the mainline Gemini app family; it's sometimes labeled "Gemini Notebook" in newer product materials from Google, though "NotebookLM" remains the name most people search for and the one Google still uses on its own support documentation. It is not a general chatbot or a live web search engine. It won't answer questions outside your uploaded sources unless you explicitly use its web-discovery feature (more on that below), and that constraint is the entire point — it trades breadth for trustworthiness. For broad research that pulls in live web sources by default, pair it with the tools covered in AI research assistants 2026.

NotebookLM does not train Google's models on the documents you upload. Google states this explicitly for both consumer and Workspace accounts — your sources stay private to your notebook unless you choose to share it, per [Google's Workspace Updates announcement](https://workspaceupdates.googleblog.com/2025/02/notebooklm-and-notebooklm-plus-now-workspace-core-service.html) on NotebookLM's enterprise data protections.

---

Step 1: Create a Notebook and Add Sources

1. Go to notebooklm.google.com and sign in with a Google account. The core product is free.
2. Create a new notebook and click Add sources.
3. Upload sources. Supported types include:
   • PDFs (papers, textbooks, reports)
   • Google Docs, Slides, and Sheets
   • Microsoft Word (.docx) files
   • Pasted text
   • Website URLs
   • YouTube videos (NotebookLM reads the video's transcript — you cannot upload a raw video file, only link to one that already has captions or a transcript available)
   • Audio files (MP3, WAV, and more than 20 other formats)
   • Images with embedded text, processed through OCR (optical character recognition — the process of converting an image of text into machine-readable text)

Tip: group related material in one notebook rather than dumping everything into a single mega-notebook. A notebook per course, per project, or per legal case keeps answers focused and citations clean. Google's own help center guidance and independent breakdowns of the product, such as DigitalOcean's NotebookLM overview, consistently point to the same lesson: NotebookLM's accuracy advantage comes specifically from a tight, relevant source set, not from cramming in everything you own.

How many sources can one notebook hold?

Limits vary by account tier. On the free (Standard) tier, you get up to 100 notebooks and 50 sources per notebook. Paid tiers — NotebookLM Plus and the versions bundled into Google AI Pro and Google AI Ultra subscriptions — raise both notebook and per-notebook source counts, up to several hundred sources on the highest tiers, according to current comparisons from Elephas and felloai. Because NotebookLM bundles pricing into wider Google AI subscriptions rather than selling as a standalone product, exact numbers shift with plan changes — check the current limits on your account before assuming a number from any single article, including this one.

---

Step 2: Ask Grounded Questions

Once sources are added, use the chat panel on the right side of the notebook. The trick is to ask the kind of question you'd ask a tutor who has read exactly these documents and nothing else:

• "Summarize the main argument of each source."
• "What does source 3 say about side effects?"
• "Where do these papers disagree?"
• "Explain Section 4.2 in plain language."
• "List every date mentioned across these documents, in order."

Every answer includes inline citations — small numbered markers you click to jump straight to the exact passage in the source document that the claim was pulled from. Always click through on anything that matters before you rely on it. This is not optional diligence; it's the entire reason to use a grounded tool instead of a general chatbot. It's the same habit we recommend for chatting with PDFs generally: AI accelerates the reading, but a human still confirms the claim.

Does NotebookLM ever get things wrong?

Yes. Grounding in your sources sharply reduces fabrication compared to an ungrounded chatbot, because the model is answering from a fixed, retrievable text rather than generating from memory. But it can still misread a passage, miss context spread across multiple documents, or misattribute a claim to the wrong source. Citation-checking is what catches these errors, which is why the workflow only works if you actually click the citations rather than trusting the summary on faith.

---

Step 3: Generate Study and Research Aids

NotebookLM can transform your sources into ready-made materials with one click, from the Studio panel:

Output	What it produces	Best for
Study guide	Key concepts, definitions, and practice questions	Exam prep, onboarding new hires
Briefing doc	An executive summary synthesized across all sources	Catching up fast on a stack of reports
FAQ	Anticipated questions with sourced answers	Reference material, team wikis
Timeline	Events in chronological order	History, project post-mortems, legal case files
Flashcards	Term-and-definition cards, with adjustable difficulty and count	Memorizing vocabulary, dates, formulas
Quiz	Auto-generated comprehension questions from your material	Self-testing before an exam
Mind Map	An interactive, branching visual map of how concepts connect	Seeing the big picture across a large source set

Table: NotebookLM's Studio outputs and what each one is for.

Flashcards and quizzes were added to NotebookLM in a rollout that started around September 2, 2025, and reached the Android and iOS apps shortly after, per Google's Workspace Updates post announcing the feature. The Mind Map feature works similarly to a concept map a student might draw by hand while studying, except NotebookLM builds it automatically from everything in the notebook and lets you click any node to jump to the source passage it came from.

These outputs are drafts grounded in your material, so they track your sources closely — but they are still a starting point, not a finished product. Proofread a generated study guide the way you'd proofread a first draft from a research assistant: check the structure is right, then verify the details.

NotebookLM's flashcard generator lets you set both the number of cards (fewer, standard, or more) and difficulty level (easy, medium, hard) before it builds the deck — a level of control most one-click AI study tools skip entirely.

---

Step 4: Create an Audio Overview

NotebookLM's best-known feature is the Audio Overview: it generates a podcast-style conversation between two AI hosts discussing your uploaded sources, in a natural back-and-forth that sounds closer to a real interview than a robotic summary readout. It's genuinely useful for:

• Reviewing material on a commute or during a walk, when reading isn't practical
• Auditory learners who retain more from listening than from reading
• Getting a friendly "explain it to me" pass over a dense document before diving into the text itself

You can steer the focus by giving the tool instructions before it generates the audio — for example, "focus on the methodology" or "spend more time on the counterarguments." The free tier includes a daily cap on how many Audio Overviews you can generate (commonly cited around three per day on the Standard tier, though check your account's current allowance since Google adjusts these limits), while paid tiers raise the ceiling substantially.

What is a NotebookLM Video Overview?

Video Overview is a newer Studio output that turns your sources into a narrated slideshow-style video, pairing generated visuals with a spoken explanation of your material. Google introduced a more advanced "cinematic" version of this feature, built on its Gemini and Veo video-generation models, with a wider rollout starting around July 29, 2025. The cinematic tier — animated video rather than a narrated slideshow — is currently reserved for subscribers on the higher Google AI Ultra plan; the standard Video Overview format is more broadly available.

Neither Audio nor Video Overview replaces reading the sources yourself for anything you'll be tested on or held accountable for. Treat them as a second pass: a way to hear or watch the material reorganized and explained before or after you read it closely, not a substitute for the reading.

---

Step 5: Bring in Live Web Sources with Discover

By default, NotebookLM only knows what you've uploaded. The Discover Sources feature (also called "Discover") breaks that boundary in a controlled way: you describe a topic in plain language, and NotebookLM searches the web, evaluates candidate sources, and recommends up to about ten that seem most relevant, each with a short AI-generated summary of why it's related to your topic, according to Google's own announcement of the feature. You add any of the recommended sources to your notebook with a single click, after which they behave exactly like anything else you uploaded manually — fully searchable, fully citable.

This matters because it changes the calculus on when to reach for NotebookLM versus a general research assistant. If you're starting from zero and need the tool to go find material for you, Discover Sources narrows that gap somewhat, though it still routes everything through your notebook and its citation discipline rather than answering off the open web the way a tool like Perplexity does. If you are weighing cloud platforms for building your own AI applications instead, see AWS Bedrock vs Google Vertex AI.

---

Who Should Use It — and How

You are…	Use NotebookLM to…
A student	Turn lecture slides and notes into study guides, flashcards, quizzes, and Audio Overviews — see how to use AI for studying
A researcher	Cross-question many papers at once and surface where they disagree or overlap
A professional	Summarize contracts, reports, and onboarding docs with clickable citations back to the original text
A teacher	Build sourced quizzes, FAQs, and summaries directly from curriculum material
A journalist or analyst	Interrogate a stack of source documents (filings, transcripts, leaked reports) without losing track of which claim came from which document

Table: Common NotebookLM users and their typical workflows.

---

NotebookLM vs. ChatGPT vs. Perplexity: When to Use Which

It helps to think of these three tools as answering three different questions. ChatGPT (and general Gemini or Claude chat modes) answers "what do you know about this topic, broadly?" — pulling from training data and, if web browsing is enabled, current pages. Perplexity answers "what does the live web currently say about this?" — it's built around real-time search with citations to public web pages. NotebookLM answers a narrower and more specific question: "what do these specific documents say?"

That narrowness is a feature, not a limitation, for a defined category of tasks. If you're studying from a syllabus, reviewing a contract, or reconciling five research papers your professor assigned, you don't want a tool that might blend in outside information you didn't ask for — you want one that sticks strictly to the material in front of you and shows its work. If you're instead trying to understand a fast-moving news topic or get a general explanation of a concept you've never studied, a broader tool with live web access or deep training knowledge will usually serve you better and faster.

Many power users run both: NotebookLM for the fixed body of source material they need to be precisely accurate about, and a general chatbot for exploratory questions, brainstorming, or anything outside the notebook's scope. For a broader survey of tools built specifically for literature review and citation tracking rather than general Q&A, see AI research assistants 2026.

---

A Realistic Workflow: Using NotebookLM for Exam Prep

Here's a concrete example of how the pieces fit together, using a student prepping for a midterm as the case study.

First, the student creates one notebook per course rather than one giant notebook for the semester. Into it goes every lecture slide deck, the assigned textbook chapters (as PDFs), and a few recorded lecture videos linked by YouTube URL. Keeping the notebook scoped to a single course, rather than mixing in material from other classes, is what keeps citations sharp — a lesson echoed across independent user guides like the DEV Community walkthrough of common NotebookLM study workflows.

Second, before touching any of the one-click Studio outputs, the student spends fifteen minutes asking direct questions in chat: "What are the three main theories covered in chapters 4 through 6?" and "Where does the textbook disagree with what the professor said in lecture 8?" This step matters because it forces active engagement with the material rather than passively accepting a generated summary — a distinction that matters for retention regardless of which AI tool is involved.

Third, once the student has a feel for the material, they generate a study guide and a set of flashcards at "hard" difficulty, then take the auto-generated quiz cold, without reviewing the flashcards first, to get an honest read on what they actually know versus what they think they know.

Fourth, on the commute to campus the day before the exam, they queue up an Audio Overview with instructions to "focus on the topics I got wrong on the practice quiz," turning a 40-minute train ride into a targeted review session instead of dead time.

This layered approach — direct questioning, then structured study aids, then self-testing, then a lower-effort review pass — gets more out of the tool than jumping straight to "generate everything and skim it," because each stage forces a different kind of engagement with the source material.

---

Tips and Limits

Get more out of it:

• Keep notebooks topic-focused for sharper, more relevant answers.
• Ask follow-up questions in chat — NotebookLM remembers the conversation within a notebook, so you can build on earlier answers instead of restating context each time.
• Use citations religiously. Never quote or rely on an answer you haven't traced back to its source passage.
• Use the Mind Map feature early in a project, before you've read everything closely, to get oriented on how the material connects.
• For group projects or shared research, share the notebook rather than exporting summaries manually — everyone works from the same grounded source set.

Know the limits:

• NotebookLM only knows your uploaded sources and, if you use Discover, whatever web pages you've explicitly pulled in. It has no access to outside facts beyond that.
• Very large or poorly scanned PDFs can lose fidelity during text extraction; clean, text-based PDFs work far better than scanned images run through OCR.
• Free-tier usage caps — daily chat queries, daily Audio Overviews, sources per notebook — are real limits, not soft suggestions, and heavy users will hit them.
• Video files cannot be uploaded directly; only the transcript of a YouTube video is used, so a video without captions or a transcript won't work as a source.
• Cinematic-tier Video Overviews currently require the top Google AI Ultra subscription tier, not the free or mid-tier plans.
• Always check Google's current data policy before uploading confidential, legal, medical, or otherwise sensitive material, even though Google states it does not train models on your uploaded documents.

---

Frequently Asked Questions

Q: Is NotebookLM free? A: Yes. Google NotebookLM has a free (Standard) tier that includes notebook creation, source uploads, chat, study aids, and a daily allowance of Audio Overviews. Google also sells paid tiers — NotebookLM Plus and versions bundled into Google AI Pro and Google AI Ultra subscriptions — that raise the notebook count, sources per notebook, and daily usage caps substantially. The free tier is fully functional for most individual users; heavy or professional use is where the paid tiers start to matter.

Q: How is NotebookLM different from ChatGPT? A: NotebookLM answers only from the sources you upload and cites the exact passage behind each claim, which makes it far more accurate to your specific material and much less prone to fabricating details. ChatGPT answers from broad training data and, when browsing is enabled, live web pages, which is more flexible for open-ended questions but carries a higher risk of inventing specifics that sound plausible but aren't true. Use NotebookLM when fidelity to a defined set of documents matters more than breadth.

Q: Can NotebookLM summarize a YouTube video? A: Yes, but indirectly. Add the video's URL as a source and NotebookLM reads its transcript rather than the video file itself, so you can ask questions about it, summarize it, or include it alongside other sources in the same notebook. It works best on videos with clear spoken content and accurate captions; a video without a usable transcript won't work as a source at all.

Q: Is it safe to upload private documents to NotebookLM? A: For routine material, generally yes. Google states that NotebookLM does not train its models on uploaded Workspace or consumer user data, and that sources stay private to your notebook unless you explicitly share it. Still, check Google's current data policy before uploading anything highly confidential, legal, or medical, and for the most sensitive documents, consider tools with explicit enterprise no-training guarantees or a local model that never leaves your machine.

Q: How many sources can NotebookLM handle per notebook? A: It depends on your account tier. The free Standard tier allows up to 50 sources per notebook and up to 100 notebooks total; paid tiers bundled into Google AI Pro or Ultra subscriptions raise both limits substantially. For large projects, splitting material across multiple focused notebooks tends to produce sharper, more relevant answers than cramming everything into one oversized notebook, regardless of your tier's technical ceiling.

Q: What's the difference between an Audio Overview and a Video Overview? A: An Audio Overview is a podcast-style conversation between two AI hosts discussing your sources, generated as audio only. A Video Overview pairs narrated explanation with generated visuals in a slideshow-style format, and a more advanced "cinematic" animated version, built on Google's Gemini and Veo models, is currently limited to the top Google AI Ultra subscription tier. Both are meant as a second pass over material you've already engaged with, not a replacement for reading the sources.

Q: Can NotebookLM make flashcards and quizzes? A: Yes. Both features were added to NotebookLM's Studio panel in a rollout that began around September 2025 and later reached the mobile apps. Flashcards let you set the number of cards and a difficulty level before generation; quizzes auto-generate comprehension questions grounded in your uploaded sources, so you can self-test on material you've actually added to the notebook.

Q: Does NotebookLM search the web on its own? A: Not by default. NotebookLM only answers from sources you've explicitly added. The Discover Sources feature is the exception: you describe a topic, and NotebookLM searches the web and recommends candidate sources for you to review and add manually. Nothing it finds becomes part of your notebook until you click to add it.

Q: Can I use NotebookLM for group projects or shared research? A: Yes. You can share a notebook with collaborators, who then work from the same grounded source set and see the same citations. This tends to work better for team accuracy than each person generating separate AI summaries independently, since everyone is answering from the identical, verifiable document base rather than potentially divergent AI outputs.

Q: Is NotebookLM the same thing as "Gemini Notebook"? A: Largely yes. Google has folded NotebookLM more tightly into its Gemini product family over time, and some of Google's own newer materials refer to it as "Gemini Notebook." The underlying tool, its source-grounding approach, and its feature set are the same regardless of which name a given page uses.

---

Read next

• Best AI PDF Tools 2026 — Chat With Your Documents
• AI Research Assistants 2026
• How to Use AI for Studying

---

Bottom Line

NotebookLM turns your own documents into a citeable research assistant: upload sources, ask grounded questions, generate study guides, flashcards, quizzes, and mind maps, and listen to or watch an Audio or Video Overview — all built around answers you can trace back to the exact passage they came from. Its core advantage over a general chatbot isn't raw capability; it's discipline. It answers only from what you give it, and it shows you where every answer came from.

Use it whenever being right about your specific material matters more than answering anything in the world, and treat the citation links as mandatory reading, not an optional extra.

Last verified: July 2026