NotebookLM Tutorial (2026 September): Sources, Audio Overviews and Study Workflows

NotebookLM is the Google AI tool that answers only from the documents you give it. You upload a syllabus, a stack of PDFs, a few lecture recordings and a YouTube playlist, then you ask questions and every answer comes back with numbered citations that jump to the exact passage. That is the whole idea: grounded answers, not a general chatbot guessing. This guide has 9 sections, 5 copy-paste blocks of prompts or commands, an 8-question Q&A, and takes about 25 minutes to read.

This NotebookLM tutorial walks you through the full workflow as the tool works in late 2026. You will create a notebook, add every source type it accepts, ask questions that cite their evidence, generate Audio Overviews and Video Overviews, turn the notebook into study guides and timelines, and share it with a study group or a team. Two complete workflows are included, one for exam study and one for business research, plus a section on limits and the errors you will actually hit.

You need a Google account and a browser. NotebookLM runs at notebooklm.google.com and also has a mobile app. No installation, no command line. If you later want to combine it with the Gemini app or an agent, the AI Mastery course has a full module on the Google AI toolset, but everything below works on its own.

Most AI chat tools answer from their training data plus whatever you paste in. NotebookLM flips that. A notebook starts empty. It knows nothing until you add sources, and once you do, its answers are restricted to those sources. Ask about something that is not in your documents and it tells you it cannot find that in the sources instead of inventing an answer.

Under the hood it runs on Gemini, the same model family behind the Gemini app. The difference is the grounding layer: every claim in a response is tied to a citation, and you can click through to verify it. That makes NotebookLM a research and study tool first, and a writing tool second.

What it is good at: summarising long material, answering specific questions across dozens of documents, finding where a topic is discussed, comparing what two sources say, and turning a pile of reading into audio you can listen to on a commute. What it is not: a web search engine, a code editor or a general assistant. If you want the open web, use Gemini with Deep Research or the Discover sources button inside NotebookLM, which we cover below.

The interface has three panels. Sources on the left, where you add and tick documents. Chat in the middle, where you ask questions. Studio on the right, where Audio Overviews, Video Overviews, study guides and notes live. Everything in this tutorial maps to one of those three panels.

Open notebooklm.google.com, click Create new and name the notebook after the project, not the date. Names like Biology 101 Final or Q4 competitor scan keep the list readable once you have twenty notebooks. Then click Add source and pick the type.

PDF and text files. Drag the file into the upload area or click to browse. Text-based PDFs work immediately. Scanned PDFs, where the pages are images, need a text layer first; run them through a free OCR tool or Google Drive's open-with-Docs conversion before uploading. Plain .txt and Markdown files are accepted too.

Google Docs and Slides. Choose Google Drive, search or browse, and pick the file. The document is imported as a snapshot. If you edit the Doc later, open the source in NotebookLM and click Click to sync with Google Drive, otherwise answers stay based on the old version.

Websites (URLs). Choose Website, paste the address, confirm. NotebookLM pulls the readable text of that single page, not the whole site. Paywalled pages, login pages and pages that render content only through heavy JavaScript often come back empty; copy the text and paste it as a text source instead.

YouTube. Choose YouTube and paste the video link. NotebookLM uses the video's transcript, so it works for public videos that have captions, auto-generated or uploaded. Private videos, live streams and videos without any transcript fail.

Audio. Upload an .mp3, .wav or .m4a lecture recording or meeting. NotebookLM transcribes it and treats the transcript as the source. Transcription takes a few minutes for a long recording; the source shows a spinner until it is ready.

Pasted text. Choose Copied text, paste, give it a title. Use this for email threads, notes, meeting minutes and anything you cannot upload as a file.

A quick way to add several at once is Discover sources. Describe a topic in a sentence and NotebookLM searches the web, proposes a list of pages, and lets you tick which to import. Review them before adding; the tool is grounded in whatever you accept, so a weak source produces weak answers.

With sources loaded, the chat panel shows an auto-generated notebook summary and a few suggested questions. Start with a broad one to check the sources imported cleanly, then go specific. A good first prompt:

Summarise each source in two sentences and flag any source that seems off-topic or empty.

Read the answer against the source list.

Every answer carries small numbered citation chips. Click one and the source opens on the right with the cited passage highlighted. This is the habit that makes NotebookLM trustworthy: you verify the claim in three seconds instead of taking it on faith. If a citation points to a passage that does not support the sentence, treat that sentence as a hallucination and rephrase your question.

Ticking and unticking sources narrows the answer. Ask a question with only chapter 4 ticked and the model cannot draw on chapter 5. Use this for comparison questions: tick two competing sources and ask where they disagree.

Prompts that work well in the grounded setting are precise about form and scope. Some examples you can paste:

List every definition given for "opportunity cost" across the sources, with the source name next to each.
What evidence do the sources give for X? Quote the strongest sentence for each point.
Create a table: concept, which source covers it, page or timestamp.
What does source A claim that source B contradicts?

Each returns citations you can check.

When you get an answer worth keeping, click Save to note. Saved notes live in the Studio panel and can later be converted into a source themselves, which is how you build a second-generation notebook out of your own synthesis. For the prompt craft behind these questions, the prompt engineering guide goes deeper.

Audio Overviews are the feature that made NotebookLM famous. In the Studio panel, click Audio Overview and then Generate. A few minutes later you get a podcast-style conversation between two AI hosts who discuss your sources, explain the key ideas, and joke a little. It is grounded like everything else: the hosts only talk about what is in the ticked sources.

Before generating, click Customize. This small text box changes the result more than anything else. Tell the hosts who the listener is, what to focus on and what to skip. For example:

The listener is a first-year nursing student revising for a pharmacology exam tomorrow. Focus on drug classes, mechanisms and the three most testable side effects per class. Skip history and anecdotes.

Then generate and listen.

Once the audio exists you can play it inside NotebookLM, download the file, or share it with anyone who has access to the notebook. There is also an interactive mode: press the join button while the overview plays and you can ask the hosts a question with your voice; they answer from the sources and then carry on. Output language follows your account's output-language setting, so change that first if you want the hosts speaking Spanish or Hebrew.

Video Overviews are the newer sibling. Same Studio panel, Video Overview, optional customisation, generate. Instead of two hosts you get a narrated slide-style video with headings, bullet points, simple diagrams and quotes pulled from the sources. It suits visual learners and it makes a decent internal explainer for a team that would never read the underlying report.

Both formats take longer than a chat answer. Queue one, keep working in chat, and come back. If a generation sits for a very long time, cancel it and retry with fewer sources ticked; a notebook with forty long sources produces a vaguer overview than one with the five that matter.

The Studio panel also has one-click outputs that write a document from the sources. Each appears as a note you can edit, copy or convert back into a source. The main ones:

Study guide. Produces short-answer questions with an answer key, a glossary of key terms, and essay-style prompts. Built for revision; it is the fastest way to generate a self-test from lecture notes.

FAQ. Turns the sources into a list of the questions a newcomer would ask, each answered from the material. Good for onboarding docs and for spotting what the sources never actually explain.

Briefing doc. An executive summary: main themes, key facts, important quotes with citations. This is the output to hand to someone who has five minutes.

Timeline. Extracts every dated event across the sources, orders them, and adds a cast of characters. Useful for history, case law, project post-mortems and any narrative with a sequence.

Mind map. An interactive tree of the topics and subtopics found in the sources. Click a node and NotebookLM asks a question about it in chat. Handy for seeing the shape of a subject before you start reading.

Flashcards and quizzes. Generate a set of question-and-answer cards or a short multiple-choice quiz from the ticked sources. You can pick difficulty and roughly how many items. Treat them as a first draft and correct any card that cites a weak passage.

Any of these can be steered with the same trick as Audio Overviews: ask for it in chat instead of clicking the button, and add constraints. For instance:

Create a study guide from sources 1 to 3 only. Twenty short-answer questions, ordered from easy to hard, with the citation after each answer.

Save it as a note when it looks right.

Here is a complete exam workflow you can copy. It assumes a course with lecture slides, a textbook PDF, a few recorded lectures and your own notes. Total setup time is about an hour, then the notebook works for you until the exam.

Step 1: build the notebook. Create Course name: exam month. Add the slides from Drive, the textbook as one source per chapter, each lecture recording as an audio source, and your notes as pasted text or a Doc. Wait for the audio transcriptions to finish before moving on.

Step 2: map the subject. Generate the mind map. Compare it with the official syllabus. Any syllabus topic that is missing from the map means your sources do not cover it, so find the material now rather than the night before.

Step 3: build the self-test. With all sources ticked, generate a study guide. Then, in chat, ask for twenty more questions on the two weakest topics from the map, with citations. Save both as notes.

Step 4: daily loop. Each day tick only that day's chapter and lecture. Generate a flashcard set. Answer every card out loud before flipping it. For any card you get wrong, click the citation and reread the passage, then ask the chat to explain it a different way.

Step 5: audio for dead time. Generate one Audio Overview per week with the customization prompt naming that week's topics and the listener as you. Play it while commuting or cooking. Use interactive mode to ask the hosts about anything that confused you.

Step 6: final week. Tick everything, ask for a timeline if the subject has a sequence, and ask for a one-page briefing doc of the most testable facts. Print it. Then run a full practice test outside NotebookLM, because the tool can only test what your sources contain; a real practice test finds the gaps your sources have.

The business version of the workflow has the same skeleton, but the sources and the outputs change. Suppose you need a competitor and market scan for a product decision.

Step 1: collect. Create Product X: market scan. Add the competitors' pricing and feature pages as URL sources, their documentation home pages, two or three analyst articles, and your own internal notes as pasted text. Add any recorded customer interviews as audio; the transcripts become searchable immediately.

Step 2: run Discover. Use Discover sources with a one-sentence topic to pull in pages you missed. Review each suggestion. Reject anything that is a content farm or a sales page dressed as research.

Step 3: structured comparison. In chat, ask:

Build a table comparing each competitor on pricing model, target customer, integrations and stated limitations. Cite the source for every cell. Leave a cell blank if no source covers it.

Blank cells are your research gaps.

Step 4: interview mining. Tick only the interview transcripts and ask for the five most repeated complaints, with a quote and a citation for each. Then ask which complaints none of the competitor pages address. That intersection is usually the most interesting slide in the deck.

Step 5: outputs. Generate a briefing doc for stakeholders, an FAQ for the sales team, and a Video Overview for the people who will never open the notebook. Customize the video with the decision you want the audience to focus on.

Step 6: keep it alive. A market scan rots fast. Set a monthly reminder to re-add the pricing pages (URL sources are snapshots) and to sync any Google Doc sources. If you later want this pipeline to run itself, the article on how to build an AI agent for beginners shows how to automate collection, and the agents module of the AI Mastery course takes it further.

Click Share in the top right of any notebook. You get the same pattern as Google Docs: add people by email as Viewer or Editor, or switch link access so anyone with the link can view. Viewers can ask questions in chat and play existing Audio Overviews, but they cannot add or remove sources. Editors can do everything you can except delete the notebook.

For a study group, one editor who owns the sources and everyone else as viewers keeps the notebook clean. Members ask their own questions and save their own notes without anyone accidentally deleting chapter 7. For a team, the same rule applies: the researcher edits, the stakeholders view.

Public link sharing is useful for a course reader or a reference notebook you want to hand out. Before flipping it on, reread the alert above about copyrighted and confidential material, and remember that a public notebook exposes the sources themselves, not just the chat.

Sharing across a personal Google account and a Google Workspace account can fail with a permissions error. Workspace admins control whether NotebookLM is enabled and whether external sharing is allowed. If a collaborator cannot open your link, ask which account they are signed into first; that solves most cases.

You can also share a single Audio Overview without sharing the notebook. Open the overview, use the share option on the player, and the recipient gets the audio alone. Downloading the file and sending it works too.

NotebookLM has caps on the number of notebooks per account, sources per notebook, words per source, file size per upload and chat queries per day. The free tier is generous for a student; the paid tiers inside Google AI Pro, Google AI Ultra and Workspace raise every cap and add more Audio Overview generations. The numbers change, so check the current NotebookLM help page rather than trusting a screenshot from a forum. Whatever the tier, the practical limit is quality: ten focused sources beat fifty vague ones.

Here are the errors people hit most, with the fix for each.

"Source could not be uploaded" on a PDF. Usually a scanned, image-only PDF or a password-protected file. Remove the password, or OCR the file (Google Drive: right-click, open with Google Docs, then export) and upload the text version. Very large PDFs: split by chapter.

YouTube link fails. The video has no transcript, is private, is a live stream or is a playlist URL. Open the video, confirm captions exist, and paste the single-video link. If there are no captions, download the audio and upload it as an audio source instead.

URL source imports as empty or as a cookie notice. The page is paywalled, needs a login, or renders content only via JavaScript. Open the page, select all, copy, and add it as Copied text. For a PDF linked from the page, download and upload the PDF directly.

Answers ignore my latest edits. Google Doc sources are snapshots. Open the source in NotebookLM and click Click to sync with Google Drive. For URL sources, delete and re-add.

Audio Overview stuck generating or errors out. Untick down to the sources that matter, clear the customization box of any instruction that asks for something outside the sources, and retry. If you have hit the daily generation cap on the free tier, wait or upgrade.

Audio Overview is in the wrong language. Open settings, change the output language, and regenerate. The customization box alone does not reliably switch language.

Collaborator gets "you need access" or a blank page. Wrong Google account, or a Workspace policy that blocks NotebookLM or external sharing. Ask them to check the account in the top-right avatar, then ask the admin if needed.

Confident answer with a citation that does not support it. Grounding is strong but not perfect. Ask again with tighter wording ("quote the exact sentence that says this"), untick noisy sources, and treat any unverifiable sentence as unusable. For how this compares with other Google AI tools, the best AI courses maps which tool each course teaches.