TL;DR
NotebookLM is Google’s source-grounded research assistant: it answers only from documents you upload and pins every claim to the exact paragraph it came from. As of June 2026 it runs on Gemini 3 (Gemini 3.1 Pro on the Pro and Ultra tiers), the free tier holds 50 sources per notebook, and Audio and Video Overviews now generate in 80+ languages. This guide takes you from a blank notebook to a working research workspace in about 20 minutes, then shows the question patterns that turn a pile of PDFs into a usable synthesis.
What NotebookLM actually is
NotebookLM only reasons over sources you give it. Ask a question and every sentence in the answer carries a citation chip; click it and you jump to the source paragraph. That grounding is the whole point: it trades the open-ended creativity of ChatGPT or Claude for traceability you can defend in a literature review or a due-diligence memo. It will not answer about the open web, and it won’t write you a poem. It will tell you where five papers disagree and quote each one.
Behind the scenes it switched to Gemini 3 in December 2025, and Pro/Ultra subscribers get Gemini 3.1 Pro, which improved its reasoning over long multi-source notebooks. The free tier still runs a fast Gemini 3 model and is genuinely capable for most research work.
Who this is for
Researchers, grad students, journalists, analysts, and knowledge workers who already have 5+ PDFs or articles on a topic and need synthesis rather than a single summary. It shines for literature reviews, due-diligence packets, competitive analysis, exam prep, and interview research. If you instead need fresh-web answers, use Perplexity; for open-ended drafting, use ChatGPT or Claude.
Pricing and limits (June 2026)
You cannot buy NotebookLM on its own. It ships as a benefit of one of Google’s AI plans, a qualifying Google Workspace plan, or a Google Cloud Enterprise license. The free tier is unusually generous, so start there and only upgrade if you hit the daily chat cap or need more than 50 sources in a single notebook.
| Tier | Price (USD/mo) | Sources / notebook | Notebooks | Chats / day | Audio Overviews / day |
|---|---|---|---|---|---|
| Free | $0 | 50 | 100 | 50 | 3 |
| Plus (Google AI Plus) | $7.99 | 100 | 200 | 200 | 6 |
| Pro (Google AI Pro) | $19.99 | 300 | 500 | 500 | 20 |
| Ultra (Google AI Ultra) | $99.99 / $200 | 500 / 600 | 500+ | 2,500 / 5,000 | higher |
Notes: Plus is bundled into the $7.99 Google AI Plus plan; Pro features (including Gemini 3.1 Pro inside NotebookLM) ride on the $19.99 Google AI Pro plan, formerly “Gemini Advanced.” At Google I/O on May 19, 2026, Ultra split into a $99.99 entry plan and a $200 high-limit plan, which is why Ultra has two SKUs. Always check the official NotebookLM plans page before paying, since these caps move.
Before you start
- Have a Google account, then visit notebooklm.google.com. No credit card is needed for the free tier.
- Curate, don’t dump. Synthesis quality tracks signal-to-noise: 10-20 strong sources beat 50 mediocre ones every time, and you stay well under the free 50-source cap.
- Name the notebook with your actual research question, not the topic. “Does indoor air quality affect cognitive performance?” beats “indoor air.”
- Paywalled HTML won’t import. The URL importer can’t bypass paywalls, so upload a PDF of the paywalled article instead.
Step by step
- Open notebooklm.google.com and sign in with a Google account.
- Create a notebook and add sources. Click “Add source.” Supported types: PDF, website URL, Google Doc/Slides, plain text, pasted text, Markdown, and YouTube links (it pulls the transcript). On the free tier you can add up to 50 sources per notebook, each up to roughly 500,000 words.
- Wait for ingestion. Usually under a minute per source; large PDFs take longer. A source preview appears in the left panel when it’s ready.
- Ask grounded questions in chat. Every answer is annotated with citation chips. Click one to jump to the exact source paragraph, and always confirm the citation actually supports the claim before you quote it.
- Generate an Audio or Video Overview. Audio Overview produces a 5-15 minute podcast-style dialog between two AI hosts; Video Overview (added in 2026) renders a narrated deep-dive with on-screen visuals. Both now support 80+ languages. Use the “Customize” prompt to focus on a specific question or set the expertise level.
- Pin key answers as Notes. Notes become referenceable in later prompts, which is how you accumulate findings across multiple sessions.
- Use the Studio tools. From the Studio panel, generate a Mind Map (click any branch to open a grounded chat on that subtopic), Reports (a briefing doc or competitive analysis), Flashcards, a Quiz, or a Data Table. These are pre-built syntheses you’d otherwise have to prompt for by hand.
Question patterns that actually surface insight
The single biggest upgrade to your output is to stop asking “summarize” and start asking for synthesis or disagreement:
- “What’s the consensus across these sources on X, and which source dissents?”
- “Where do these sources disagree on Y, and what’s the strongest evidence each one cites?”
- “List every claim about Z with its source and the author’s stated confidence level.”
- “Which sources contradict the most-cited claim, and on what grounds?”
- “Build a data table comparing each source’s method, sample size, and headline finding.”
Summaries are bland; disagreement maps and evidence tables are the reason to use a grounded tool at all.
A realistic first-pass workflow
Gather 10-20 curated sources, then upload them. Ask “What’s the consensus on X?” and drill into the citations to verify. Pin the three best answers as Notes. Ask “Where do these sources disagree, and what evidence backs each side?” and use that map to plan your follow-up reading. Generate a Mind Map to see the structure, then an Audio Overview to review on a commute or hand to a teammate. Budget 20-30 minutes for the first pass and return as you add sources.
FAQ
- Is my data used to train Google’s models? Per Google’s current policy, your uploads, queries, and chats in personal NotebookLM are not used to train models, and human reviewers don’t see them. Workspace/Enterprise accounts carry their own data terms. Confirm in your account’s privacy settings.
- How many sources can I add? 50 per notebook on the free tier, 100 on Plus, 300 on Pro, and 500-600 on Ultra (as of June 2026). Each source can be hundreds of pages.
- What model powers it? Gemini 3 since December 2025; Pro and Ultra subscribers get Gemini 3.1 Pro inside NotebookLM, released February 19, 2026.
- Can I share a notebook? Yes. Use the share button to invite viewers, who can read the sources and chat with them. Handy for passing a research packet to a colleague.
- Are Audio Overviews still English-only? No. Audio Overviews and Video Overviews now generate in 80+ languages; set the output language in the generation panel or in Settings.
- Can I export an Audio Overview? Yes, download it as an MP3. Adjust length (Shorter/Default/Longer) and steer the hosts in the customize panel before generating.
Common mistakes
- Uploading one source. That throws away the cross-document value and leaves you with a clunky chat over a single PDF.
- Ignoring citation chips. They’re the trust signal. An uncited sentence in an answer is a hallucination warning.
- Treating it like ChatGPT. It’s grounded, not generative; it won’t brainstorm freely or write creative prose well.
- Dumping low-quality sources. Synthesis degrades with noise. Curate to 10-30 strong sources instead of maxing out the 50-source cap with filler.
- Only asking “summarize.” Consensus, disagreement, and evidence questions return far more useful output.
- Skipping Notes. Without them you lose the ability to build on earlier sessions.