NotebookLM: Google's AI research tool for your own source collection

This article is an overview based on vendor information, documentation and publicly available reporting, not a scored test of our own. Prices and features change quickly; the vendor's own terms always take precedence. We only score what we have used ourselves for weeks: those tools are listed under AI tools compared.

What NotebookLM is, and how it differs from a chatbot

NotebookLM turns the chatbot principle on its head. You first create a notebook and fill it with your own sources: PDFs, Word files, Google Docs and Slides, web pages, public YouTube videos, audio files or pasted text. Only then do you ask questions, and the model answers solely from that material, each statement with a numbered citation that jumps to the relevant passage on click.

That is the real difference from Gemini, Claude or Perplexity: a general assistant draws on its training knowledge and usually the web as well, NotebookLM on what sits in the notebook. If you want to know what your contract says, rather than what contracts of that kind usually contain, this is the right place.

Since 16 July 2026 the product has, according to Google, been called "Gemini Notebook". The address, existing notebooks and prices remain unchanged according to the vendor; among the additions is a feature that lets the tool execute code and analyse data drawn from the sources. In this article we stick with the established name.

Typical uses: dossiers, contracts, literature, onboarding

The first case is the dossier: tender documents, an annual report, a stack of market studies. Instead of skimming 300 pages, you ask about deadlines, responsibilities or contradictions between the documents and get an answer with references that can be checked in minutes.

The second case is contracts and rulebooks. A notebook holding a framework agreement, terms and conditions and amendments answers questions such as "Which notice periods apply to which service?" considerably faster than full-text search. The cited passage remains what counts, not the summary.

Third and fourth cases: academic literature and onboarding. For a literature review, dozens of papers can be queried by method and result. And a notebook containing process descriptions, an org chart and the product manual becomes a first port of call for new staff, who can ask questions without interrupting someone every time. More examples of such tasks are in our article on AI in everyday office work.

Audio Overviews, Mind Maps and the limits

NotebookLM became known for its Audio Overviews: from the sources it produces a conversation between two synthetic voices that sounds like a podcast. Since 2025 this has been available, according to the vendor, in more than 50 languages including German; the output language is set in the settings. Added to that are Mind Maps, which show the sources' topics as a tree, plus video overviews, flashcards, quizzes and reports.

The audio formats are suited to getting into a topic on the commute, less so as a working basis. A finished episode cannot be corrected in one spot afterwards, only regenerated, and the German version sounds audibly less natural than the English one. User reports also describe cases in which the hosts invented content that was not in the sources.

The same applies in weaker form to the chat: tying answers to the sources cuts the number of fabricated statements considerably but does not eliminate them. Check the citations before an answer goes into a document. Google sets technical limits at around 500,000 words or 200 MB per source and, on the free tier, 50 sources per notebook; scanned PDFs without a text layer and web pages behind a paywall cannot be imported.

Pricing: free, Plus, Pro, Workspace, Enterprise

The basic version is free with a personal Google account and, according to the vendor as of September 2026, limited to 50 chat queries and three Audio Overviews per day. That is enough for occasional use; anyone working with it daily hits the limit.

Higher quotas are not sold as a separate subscription but come with the Google AI plans: Google AI Plus at around 5 euros a month raises the limits to 100 sources per notebook and 200 queries a day, Google AI Pro at around 22 euros to 300 sources, 500 queries and 20 Audio Overviews. The paid tiers also offer advanced sharing features and access to more capable models.

For companies the usual route is the Google Workspace subscription, whose business plans include NotebookLM. Above that sits NotebookLM Enterprise, which runs in your own Google Cloud project and is priced through Cloud sales; Google does not publish a list price for it.

Data, the GDPR and where processing happens

The most important question first: according to Google, content in NotebookLM is not used to train its foundation models unless you explicitly submit feedback. For Workspace accounts Google goes further and states that uploads, questions and answers are neither reviewed by humans nor used for training.

So the account is what matters. With a personal Google account the general terms of service apply, there is no data processing agreement, and business documents containing personal data have no place there. With a Workspace account the Cloud terms apply including a DPA, and administrators can enable or block the service centrally. What else to check in that review is in our article on AI tools and the GDPR.

On the processing location Google's documentation is reticent. The vendor guarantees storage within the EU only for NotebookLM Enterprise with the corresponding region setting, and even there the commitment, per the Cloud documentation, covers sources and chat, not Studio features such as Audio Overviews or Mind Maps. A notebook of contracts and a notebook of personnel files are therefore two very different undertakings.

Who NotebookLM is worth it for

Anyone who regularly works with a fixed stack of documents: consultancies, law and tax firms with a Workspace account, researchers, editorial teams, project leads with thick tender files. The citations make the results verifiable, and getting started costs nothing. If you already work in Google Docs you gain further, because linked documents update automatically.

It is less worthwhile for questions without a source collection of your own, where the chatbots in our category are the better tool, and for teams whose knowledge already lives in another system; there Notion AI is often the shorter route. And anyone wanting to upload confidential company documents with a personal account should hold off until a Workspace or Enterprise account is in place.

In short: NotebookLM answers only from your own sources and cites every statement, free with a personal account, with higher quotas via Google AI Plus and Pro, and via Workspace for companies. Check the citations, expect compromises in German Audio Overviews, and keep business data to Workspace or Enterprise accounts.