NotebookLM is Google's research assistant that answers only from the documents you give it. You upload PDFs, web pages, YouTube videos or audio files, then you ask your questions. Every sentence of the answer carries a clickable citation pointing to the exact passage in the source. The tool remains free with a plain Google account.
One clarification before going further: the product changed its name. Google rebranded it Gemini Notebook in July 2026. Both names refer to the same product, and this tutorial uses them interchangeably because both are still in circulation. Here is the full walkthrough, the supported formats, the real free-tier quotas, the limits nobody mentions, and what to know before putting company documents in it.
What exactly is NotebookLM?
NotebookLM is a synthesis tool grounded in your own sources. You create a space, called a notebook, drop up to fifty documents into it on the free tier, and a Gemini model answers your questions using only those documents. It never fills gaps with what it knows about the world.
That is the fundamental difference with a general assistant. Ask an off-topic question to ChatGPT and it will attempt an answer. Ask the same question to NotebookLM and it will state that the information is not in the supplied sources. That refusal is the tool's main feature.
The underlying mechanism has a name, retrieval-augmented generation. Documents are split into passages, indexed, and the ones closest to your question are passed to the model. It is the same principle described in our article on RAG and business context, except that Google ships it ready to use, with no code at all.
Why NotebookLM is now called Gemini Notebook
Google announced the rename in July 2026 on its Workspace updates blog. The announcement states that the product stays a standalone tool and nothing needs migrating, with shared notebooks and existing links still working thanks to automatic redirects. Administrators were asked to do nothing.
The check takes two seconds. The notebooklm.google address now answers with a permanent redirect to notebook.google, and the app itself redirects from notebooklm.google.com to notebook.google.com. The home page title reads Gemini Notebook. If you kept a bookmark, it still works.
The rename came with a technical addition: every notebook now gets an isolated execution environment that can write and run code to analyse your numerical data. In practice this extends the tool to tabular files, where it used to stop at text.
Creating your first notebook in four moves
Getting started takes a few minutes. Nothing to install, everything happens in the browser with a Google account.
- Open notebook.google and create a notebook. Give it a clear project name rather than a generic label, because notebooks do not talk to each other.
- Upload your sources. Drag your files in, paste URLs separated by a line break, or pick from your Google Drive. Drive sources resync every two to three minutes.
- Ask your questions in the Discussion panel. A precise question beats a request for a summary. Name the document title when you want to target one source.
- Open the Studio to produce a deliverable. Audio overview, report, mind map, data table: this is where the synthesis becomes a reusable object.
One habit to build from the first notebook: save the answers you need as a note. The discussion panel keeps no permanent history, and a useful synthesis disappears if you close the space without saving it.
What you can upload, and the two ceilings to know
The range of supported formats is the tool's real selling point. Google's official help centre lists PDFs, Word, PowerPoint, text, Markdown, CSV and ePub files, plus Google Docs, Slides and Sheets. Images, audio files, web pages, public YouTube videos and plain copy-paste are supported too.
Three restrictions hide inside that list and deserve attention before you lose half an hour. A Google Slides deck is truncated beyond one hundred slides. A Google Sheets file is capped at one hundred thousand tokens, a modest volume for a database export. And a YouTube video with no automatic transcript yields no usable text.
The Studio, or turning your sources into deliverables
The Studio is the panel that separates NotebookLM from a plain question-and-answer engine. It builds nine kinds of object from the same sources, and most of them export to Google Docs or Google Sheets.
The audio overview, in four formats
The audio overview turns your documents into a listenable conversation. Four formats exist: a deep dive with two hosts, a single-voice briefing under two minutes, a constructive critique and a formal debate. The output language is picked in the generation panel, and more than eighty languages are supported.
This is the format that works best on a commute. Take a dozen articles about your market, generate a briefing, and you have a two-minute sector review without reading anything. The trade-off is covered further down, and it is a serious one.
Deep Research, when you have no sources yet
Deep Research inverts the tool's logic. Instead of starting from your documents, it starts from an instruction, goes looking for sources on the web and drops them into your notebook. The phrasing that works is explicit about the kind of evidence expected. Ask for ten authoritative sources on a market, prioritising market sizing, user reviews and customer cases.
Pairing NotebookLM with a search engine
NotebookLM has one weakness upstream: it does not find the right sources for you, unless you spend your monthly Deep Research quota. A conversational search engine does that sorting much faster, and handing over between the two tools is the most effective working method I have found.
The sequence has three steps. You run the search in a conversational engine, forcing the document type with a search operator and bounding the date. You open the results, discard the promotional pages and keep the studies. You pour the selected links into a notebook, and only then does the analysis begin.
I filmed the whole manoeuvre in my video on cutting research time, and I wrote it up in my newsletter edition on NotebookLM. The point of this setup is a sustained conversation over a stable corpus, where a search engine fires a fresh query at every question and loses the thread.
That split by use comes from a one-pager I published on LinkedIn in January 2026. It says the essential thing for anyone starting out: no tool covers the whole cycle, and the useful question is which tier your current need sits on. To pick the general assistant that completes this setup, our overview of AI assistants compares the seven main ones.
The real limits of NotebookLM
A tutorial that only praises its subject is useless. Here are the six limits Tandem genuinely runs into on client work, in decreasing order of annoyance.
- Notebooks are siloed. Google's help centre is explicit, a notebook cannot access another one's information. No cross-search over your document base, so no unified company memory.
- The chat keeps no permanent history. An answer not saved as a note is lost when you close the space.
- Some uploads fail with no explanation. A message flags the failure, you delete the source and start over, sometimes in a different format.
- Exports do not come back. A report sent to Google Docs becomes a standalone document. Your edits never flow back into the notebook, and sharing permissions do not carry over either.
- Long-form writing escapes it. The tool extracts and structures, it does not write in your voice. To produce the final deliverable, move back to a general assistant.
- Phantom learning is waiting for you. This is the most serious limit, and it comes from how you use the tool.
Free, Plus, Pro or Ultra: what does the plan change?
The free tier covers the vast majority of individual uses. Paid plans unlock no major exclusive feature, they loosen quotas and open cinematic video plus watermark removal.
| Plan | Price per month | Sources per notebook | Questions per day | Audio overviews | Deep Research |
|---|---|---|---|---|---|
| Standard | Free | 50 | 50 | 3 a day | 10 a month |
| Google AI Plus | 4.99 euros | 100 | 200 | 6 a day | 3 a day |
| Google AI Pro | 21.99 euros | 300 | 500 | 20 a day | 20 a day |
| Google AI Ultra | From 99.99 euros | 500 to 600 | 2,500 to 5,000 | 100 to 200 a day | 75 to 200 a day |
The amounts are taken from the public Google AI plans page in October 2026, in euros and for France. The ceilings come from the help page on usage limits.
The right call depends on a single number, the count of sources per matter. Below fifty documents per subject, stay on the free version. Above it, on a due diligence or a regulatory review for example, the Pro plan earns its keep. An eligible Google Workspace licence opens the same limits with no extra individual subscription.
Are your documents confidential in NotebookLM?
This is the question that blocks adoption in companies, and the answer depends entirely on the account you sign in with. The two regimes do not offer the same guarantees.
With a personal Google account, Google states that your data is not used to train Gemini Notebook, unless you send feedback. In that case the full context of the exchange may be reviewed, meaning your prompts, your uploads and the model's answers. Simply clicking the thumbs-up on an answer therefore opens that door.
With a Google Workspace or Workspace for Education account, the guarantee is stronger and unconditional. Your uploads, prompts and the model's answers are never reviewed by humans, even if you send feedback, and never used to train the models. Google Cloud and Workspace also open enterprise-grade data protection.
This distinction deserves to be written down in your usage charter. It belongs beside the roles and sign-offs described in our article on AI governance in companies.
Should NotebookLM be part of your working day?
Yes, on a precise perimeter, and no as your main assistant. NotebookLM is the best free tool available for questioning a closed corpus with verifiable citations: a tender, a regulatory bundle, a library of meeting notes, a market to map. On that ground, no general assistant offers the same traceability.
It replaces neither your writing assistant, nor your search engine, nor a company knowledge base, since its spaces do not talk to each other. Tandem positions it as a complement inside AI literacy programmes, never as a single tool.
To go further on business uses, our three detailed use cases show how to produce a sourced pitch, formalise internal procedures and map a competitive landscape. And if slide production interests you, our comparison of AI slide generators places the Studio against the specialised tools.
Installing these habits in a team takes more than a demo. That is the point of our AI literacy programmes, built around the tools your teams already use. The first workshop always starts from the same observation, taken from our guide on AI literacy in companies. A habit takes longer to take root than a tool.
