NotebookLM, Explained
The One-Line Answer
NotebookLM is Google's AI research assistant that reads your documents — not the internet — and answers questions, finds patterns, generates content, and synthesises insights exclusively from sources you upload. Zero hallucination risk from outside data. Every answer is grounded in what you gave it.
What Makes It Different
Most AI tools draw from a giant blob of internet data. NotebookLM is source-locked: it only knows what you feed it. This means your analysis stays private, your outputs are traceable to real citations, and the model can't drift into fabricated context.
Who It's Built For
Researchers, consultants, content creators, analysts, lawyers, marketers — anyone who needs to process dense information fast. If you spend time reading documents so you can think about them, NotebookLM does the reading and hands you the thinking.
The Core Loop
Upload sources → Ask questions → Get cited answers. Add more sources to the same notebook and it synthesises across all of them simultaneously. This is where it becomes genuinely powerful: pattern recognition across dozens of documents at once.
Free vs Plus
Free tier: 100 sources per notebook, 50 notebooks, 50 audio overviews/day. NotebookLM Plus (paid): 500 sources per notebook, 200 notebooks, 500 audio overviews/day, plus customisable audio and notebook sharing.
| Capability | Free | Plus | Notes |
|---|---|---|---|
| Sources per notebook | 100 | 500 | PDFs, URLs, YouTube, text, audio |
| Notebooks | 50 | 200 | Each notebook is isolated |
| Audio overviews / day | 50 | 500 | Two AI hosts discuss your sources |
| Custom audio instructions | ✗ | ✓ | Change tone, focus, format |
| Notebook sharing | ✗ | ✓ | Collaborate with team |
| Max source size | 500,000 words / 200MB PDF | Per source limit | |
Setup, Sources & Structure
Go to notebooklm.google.com
Sign in with a Google account. Click + New Notebook. Give it a meaningful name — you'll build many of these. Think of each notebook as a dedicated research project or content operation.
Add Your Sources
Click Add Source in the Sources panel on the left. You can upload files, paste a URL, add a YouTube link, paste raw text, or connect a Google Doc or Slide. Mix and match freely — a notebook can have PDFs, video transcripts, and website content simultaneously.
Wait for Indexing
Each source takes 10–60 seconds to process. You'll see a loading indicator. Don't start prompting until all sources show a checkmark — partial indexing gives partial results. Large PDFs (100+ pages) can take 2–3 minutes.
Select Which Sources to Query
By default all sources are active. You can toggle individual sources on/off using the checkboxes in the Sources panel. This lets you compare one source vs another, or isolate a single document for deep analysis.
Start With the Auto-Generated Summary
After sources load, NotebookLM generates a notebook guide automatically — key topics, suggested questions, and source summaries. Read this first. It reveals what the model found most salient and gives you a launchpad for deeper prompting.
PDF Files
Best for: research papers, reports, ebooks, contracts. Works on text-based PDFs. Scanned image PDFs may have reduced accuracy. Keep under 200MB. Split very large PDFs into logical chapters for better results.
YouTube Links
Paste any public YouTube URL with auto-generated captions. NotebookLM imports the full transcript. Works on interviews, lectures, podcasts uploaded to YouTube, tutorials. No captions = no import.
Website URLs
Paste public web pages — articles, documentation, landing pages, blog posts. NotebookLM scrapes the visible text. Paywalled content, login-required pages, and heavy JS sites may fail to load properly.
Google Docs & Slides
Connect directly from Google Drive. Ideal for internal docs, client briefs, strategy decks. The source stays live — but NotebookLM takes a snapshot at the time of upload, so updates to the doc aren't automatic.
Audio Files
Upload MP3/WAV recordings. NotebookLM transcribes and indexes. Useful for recorded interviews, meeting recordings, voice memos. Accuracy depends on audio quality — clean recordings work best.
Copied Text
Paste raw text directly. Use for transcripts you've already pulled, email threads, Slack exports, or any content that doesn't fit another format. No file needed — just paste and go.
Everything NotebookLM Can Do
Chat Interface
Ask any question. Every answer includes inline citations linking back to the exact passage in your source. Click any citation to jump to the original text. The model only draws from your uploaded sources.
Notes Panel
Save any AI response as a note. Write your own notes. Pin key insights. Notes become sources themselves — you can ask NotebookLM to synthesise across your saved notes as a second pass of analysis.
Audio Overview
Two AI hosts have a deep-dive podcast conversation about your sources. 10–15 minutes of generated audio. Surprisingly good for grasping complex topics quickly while commuting or exercising.
Notebook Guide
Auto-generated on upload. Includes topic summaries, key questions to explore, and a structured overview. Think of it as the model's first read — tells you what it found significant.
Multi-Source Synthesis
With multiple sources loaded, ask cross-cutting questions. "What do all three sources agree on?" or "What does Source A say about X that Source C contradicts?" This is the real competitive advantage.
Export & Share (Plus)
Share a notebook with collaborators. They can query the same sources. Plus tier enables custom sharing permissions. Great for research teams, client briefings, or content teams working from a shared source library.
Save AI responses as notes, then ask NotebookLM to "synthesise all my saved notes into a final executive summary." You're using your own curated outputs as the new source layer — massively improves final output quality.
⚠️ No Real-Time Web Access
It does not browse the internet during a query. If you want current data, scrape the URL first and paste it as a source — then query it.
⚠️ Paywalled Content
It cannot access content behind logins. Workaround: copy-paste the text directly as a source, or use your own downloaded PDF.
⚠️ Image-Heavy PDFs
If a PDF is mostly scanned images (no text layer), extraction is unreliable. Use an OCR tool like Adobe Acrobat first to create a text-searchable PDF.
⚠️ Private YouTube Videos
Only public YouTube videos with auto-generated captions work. Unlisted videos may or may not load. Private videos will fail entirely.
PDF & Document Prompts
Eight precision prompts for extracting maximum intelligence from any document. Upload your PDF, then fire these exactly as written.
Analysis Prompts
YouTube Channel Mining Prompts
Add 10–50+ video URLs from one or more channels as sources. NotebookLM imports every transcript. Now you have the full intellectual DNA of that channel indexed and queryable. Fire these prompts to reverse-engineer any creator's system.
Channel Intelligence Prompts
Audio Overview Mode
What Is Audio Overview?
Click Generate on any notebook and NotebookLM creates a 10–15 minute podcast-style conversation between two AI hosts who discuss your sources. They debate ideas, ask each other questions, and summarise insights — all grounded in your uploaded content. It's genuinely the fastest way to absorb complex material.
Power Tips & Pro Moves
📌 Notes as Second-Layer Sources
Save every strong AI response as a note. Then prompt: "Synthesise all my saved notes into a final strategic summary." You're using curated outputs as inputs — massively improves coherence and quality of final deliverables.
🔗 Isolate Sources for Comparison
Uncheck all but two sources and ask: "What does Source 1 argue that Source 2 explicitly contradicts?" Perfect for comparing competing frameworks, vendor proposals, or research with conflicting conclusions.
📁 Build Topic Notebooks
Don't mix your AI research notebook with your legal contract notebook. Keep one notebook per project or topic. Clean separation means cleaner outputs — no cross-contamination from unrelated sources.
🎯 Specify Your Role in the Prompt
Always tell NotebookLM who you are: "As a B2B marketing consultant advising a law firm..." The model calibrates its output to your context. Generic prompts give generic answers.
🔄 Iterative Prompting Chain
Don't try to get everything in one prompt. Start broad → save the response → then prompt: "Take finding #3 from your last response and go three levels deeper." Layer your analysis progressively.
📊 Ask for Tables By Default
Append "Format as a markdown table" to any extraction prompt. Tables are copy-pasteable into Notion, Sheets, or your CRM instantly — saves massive reformatting time downstream.
🎙️ Use Audio to Pre-Warm
Generate the audio overview first, listen on a walk, then return for deep prompting. You arrive at your session already oriented — your questions become sharper and your prompts more targeted.
🧠 Add Your Own Notes as Sources
Paste your existing thinking, observations, or hypotheses as a text source. Now ask NotebookLM to "challenge my assumptions using evidence from the other sources." Your brain vs. the data.
⚡ The 5-Source Stack
For any competitive research project: load 3 competitor transcripts + 1 industry report + 1 customer review scrape. Then fire the Gap Analyst prompt. This combination almost always surfaces non-obvious opportunities.
Always Assign a Role
Start every prompt with Act as a [role]. "Act as a senior analyst" vs "Act as a skeptical investor" produces wildly different outputs from the same document. The role shapes the lens.
Specify the Output Format
Tell it exactly what you want: markdown table, numbered list, three-paragraph briefing. If you don't specify, it picks a format that may not suit your workflow.
Add a Length Constraint
Specify: in under 200 words or a 5-minute read. Without this, responses can sprawl. Constraints force the model to prioritise — and often produce tighter, higher-quality outputs.
End With a Decision Question
Close your analysis prompts with: "Based on this, what is the single most important thing I should do first?" Forces the model to prioritise and give you an actionable bottom line.