NotebookLM turns a pile of documents into something you can question. Upload PDFs, docs, links, and videos, then chat with an AI that answers only from those sources, with citations you can click. In 2026 it also generates podcast-style audio, narrated video explainers, and mind maps from whatever you feed it. If you want to see how far the research side goes, we covered its agentic mode in our Deep Research breakdown.
This guide walks through the whole tool: setup, source limits, the features worth your time, and what the paid tiers change. There’s also a section for developers, because NotebookLM has become a quiet superpower for digesting API documentation, and it pairs well with a platform like Apidog where those API specs get designed and tested in the first place. Everything below reflects the product as of August 2026, verified against Google’s official support documentation.
What is NotebookLM and is it free?
NotebookLM is Google’s source-grounded AI research assistant. Yes, it’s free: the standard tier costs nothing and includes 100 notebooks, 50 sources per notebook, 50 chat queries per day, 3 Audio Overviews per day, 3 Video Overviews per day, and 10 daily generations of reports, flashcards, quizzes, and mind maps. You need a Google account and a browser, or the mobile app on Android or iOS.
“Source-grounded” is the part that matters. A regular chatbot answers from training data and the open web, which means it can drift or fabricate. NotebookLM answers from your uploaded sources with inline citations for every claim, and if your sources don’t cover a question, it says so instead of guessing. That design makes it reliable in a way general chatbots aren’t.
One naming note: Google has been folding NotebookLM into the Gemini brand during 2026, and official support pages now call it Gemini Notebook. Same product, same limits, new label. The app still lives at notebooklm.google.com, and everyone still calls it NotebookLM, so this guide does too.
Setting up your first notebook step by step
The setup takes about two minutes:
- Go to notebooklm.google.com and sign in with any Google account. No waitlist, no credit card.
- Click “Create new notebook.” A notebook is a project container: one research topic, one course, one API, one deal. Keep them focused; the AI answers from everything in the notebook, so mixing unrelated topics muddies responses.
- Add your first sources. The upload panel accepts PDFs, Google Docs and Slides, website URLs, YouTube links, audio files, Markdown, and pasted text. Drag in three or four documents to start.
- Wait for processing. NotebookLM indexes each source in seconds to a couple of minutes depending on size, then writes an automatic summary of the notebook.
- Start asking questions. The chat panel sits in the middle. Ask something specific, check the citation numbers in the answer, and click one to jump to the exact passage in the source.
- Save useful answers as notes. Pin any response to the Studio panel so it survives beyond the chat session.
That’s the core loop: add sources, interrogate them, save what’s useful. Everything else in the product builds on it.
Adding sources: what works and where the limits are
Source quality decides answer quality, so it’s worth knowing the boundaries.
What you can add:
- PDFs and text files, including scanned documents
- Google Docs and Google Slides straight from Drive
- Public website URLs, which NotebookLM scrapes as text
- YouTube videos with captions; it reads the transcript, not the pixels
- Audio files such as lecture recordings or meeting audio
- Markdown files and copied text blocks
The limits that bite: the free plan caps you at 50 sources per notebook, and individual sources have long carried a ceiling of roughly 500,000 words or 200 MB per file. A 900-page PDF fits. Fifty of them fit in one notebook. That’s more headroom than most projects need.
What trips people up: paywalled pages and login-gated sites won’t scrape, so save them as PDFs first. YouTube videos without captions give the model nothing to read. And a stale source stays stale; if the underlying Google Doc changes, click the sync button on that source to pull the update, because it won’t refresh on its own.
The features worth using
NotebookLM ships more features than most people touch. Four earn their place.
Audio Overviews. The headline act. Two AI hosts discuss your sources in a podcast format that sounds unsettlingly human, and interactive mode lets you interrupt mid-episode to ask questions. You can steer the episode before generating: focus on one chapter, skip the background material. Free users get 3 per day.
Video Overviews. Narrated slide-style explainers generated from your sources, with selectable visual styles. They’re slower to generate than audio but land well when you need to share an explanation with someone who won’t read the source material. Also capped at 3 per day on the free tier.
Mind maps. One click produces a branching diagram of the concepts across all your sources. This is the fastest way to survey a topic you don’t know yet: generate the map, spot the branch you don’t understand, click it, and ask questions.
Reports, flashcards, and quizzes. The Studio panel turns sources into briefing docs, study guides, and self-test quizzes. Students get the obvious use; for work, the briefing doc is an underrated way to prep for a meeting about a topic you inherited yesterday. These share a pool of 10 generations per day on the free plan.
Under the hood, all of this runs on Gemini 3 models in 2026, and every tier gets the same model. Paying doesn’t buy you a smarter NotebookLM, only a bigger one.
Free vs paid: what the upgrade buys
Google sells NotebookLM’s upper tiers through its Google AI subscriptions, not as a standalone plan. The differences are volume, not intelligence:
| Limit | Free | Google AI Plus | Google AI Pro | Google AI Ultra |
|---|---|---|---|---|
| Notebooks | 100 | 200 | 500 | 500 |
| Sources per notebook | 50 | 100 | 300 | 500 to 600 |
| Chat queries per day | 50 | 200 | 500 | 2,500 to 5,000 |
| Audio/Video Overviews per day | 3 | 6 | 20 | up to 200 video |
Pro also removes the watermark from generated videos in most regions, and Ultra unlocks the cinematic Video Overview style built on Google’s Veo video model. Google AI Pro runs $19.99 per month, with the Plus tier reported at $4.99; students in several countries have gotten Google AI Pro free for a year, which is the cheapest legitimate route to the higher caps.
The honest advice: stay on free until a limit stops you. The 50-chat daily cap is the one most heavy users hit first, and only then does Plus or Pro make sense.
The developer workflow: interrogating API documentation
Here’s the use case that earns NotebookLM a spot in a developer’s toolbar. API documentation is exactly the source material NotebookLM handles well: long, structured, boring, and full of details you need at unpredictable moments.
Try this setup. Create a notebook for the API you’re integrating. Load the provider’s reference docs as URLs or PDFs, add the relevant RFCs (OAuth 2.0’s RFC 6749 is a natural fit), and drop in your own OpenAPI specification exported as Markdown or a shareable doc. Then ask real questions: “What’s the rate limit on the batch endpoint?”, “Which scopes does the refresh flow require?”, “Do any endpoints in our spec return a 429 without a Retry-After header?”. Answers come back with citations into the exact section of the spec, which beats tab-cycling through documentation sites. Teams working with Gemini’s own API can load Google’s reference docs the same way and query model parameters instead of scrolling for them.
Know the boundary, though. NotebookLM reads documents; it never calls an endpoint. It will faithfully cite a response schema that your server stopped honoring two releases ago. The spec is a promise, and something has to check the promise against production. That’s where Apidog picks up: it holds the OpenAPI spec as the living source of truth, runs the real requests, validates responses against the schema, and catches the drift a document reader can’t see. Use NotebookLM to understand an API and Apidog to verify it, then re-export the validated spec back into your notebook whenever it changes.
FAQ
Does NotebookLM train on my uploaded data? Google states that NotebookLM does not use your sources, queries, or responses to train its models, for both free and paid personal accounts. Workspace accounts get contractual data protections on top. Treat genuinely sensitive material with normal caution anyway, since policies evolve.
Can I share a notebook with other people? Yes. Notebooks share like Drive files, with viewer and editor roles, and viewers can chat with the sources without being able to change them. Public share links to Audio Overviews work too, which is how those AI podcast episodes spread.
Is there a NotebookLM API I can build on? Not a public one as of August 2026. If you want programmatic source-grounded answers, the practical route is building on Gemini models directly; a self-hosted option like Gemini 3 Pro through Ollama covers the local end, and Google’s File API covers uploads in the cloud.
What languages does it support? Chat and outputs work in dozens of languages, and you can set the output language independently of your sources. Audio Overviews now speak a wide set of languages, though non-English voices trail the English hosts in polish.
Wrapping up
NotebookLM is the rare AI tool where the free tier is the whole product: 50 sources per notebook, cited answers, audio and video generation, and mind maps, at zero cost. Set up one notebook around a real problem this week; our three real-world scenarios post is the natural next read once the basics click.
And if the documents you’re loading are API docs, close the loop. NotebookLM tells you what the documentation says; Download Apidog to design the spec, mock it, test it, and make sure the documentation stays true.
