TL;DR
Google's Gemini Notebook (formerly NotebookLM) turns your own PDFs, docs, links, and videos into a research assistant that answers only from your material — with citations pointing at the exact source. It's free with a Google account and takes about ten minutes to set up. Here is the setup, what it can generate for you, and the limits worth knowing.
What it actually is
Gemini Notebook is Google's source-grounded AI research tool. You feed it your documents and it answers questions using only that material, showing a citation for every claim so you can click through to the exact passage it used. It does not browse the open web for answers, which is what keeps it from inventing facts the way a general chatbot can.
The tool launched as NotebookLM and Google renamed it Gemini Notebook on July 16, 2026, moving it to notebook.google.com (old links redirect). It runs on Gemini 3 in the browser. The free tier is generous: 100 notebooks, 50 sources per notebook, and each source can hold up to 500,000 words or 200MB. Most people never need to pay.
Set it up in ten minutes
Go to notebook.google.com and sign in with your Google account. Click Create, and give the notebook a name tied to one project — a notebook per project is the organizing principle, because notebooks are independent and the tool cannot query across two notebooks at once.
Then add sources: click Add Source and upload PDFs, text files, or audio, pull in Google Docs and Sheets from your Drive, paste web links or YouTube URLs, or paste text straight from your clipboard. Wait for the files to finish processing — the tool scans and indexes them automatically, no configuration needed.
Make it do the work
The core move is the chat box: ask questions about your material and every answer comes back with numbered citations to your sources. Ask it to compare two documents, pull every deadline out of a contract stack, or explain a dense paper in plain language.
Then open the Studio panel. It can generate an Audio Overview — a podcast-style two-host conversation about your sources you can listen to on a walk — plus video overviews, mind maps, flashcards, quizzes, study guides, and written reports. There is also a Deep Research mode for longer multi-step investigations. Pin anything useful as a saved note so it survives between sessions.
The rules that make it good
Garbage in, garbage out: the answers are only as good as the sources you upload, so keep each notebook focused and drop anything off-topic. One notebook per project, and remember you cannot search across notebooks — split big efforts deliberately.
Two honest limits. First, your files are processed on Google's servers, so keep confidential material out. Second, the free usage budget refreshes every five hours, so marathon sessions can hit a wall — plan accordingly. And as with any AI output, verify anything that actually matters against the cited source before you act on it.
The real test: a family's medical records
Here's what this looks like when it matters. I uploaded my mom's complete cancer treatment history — every record, every doctor visit recording, every medication — into a shared notebook the whole family can access.
Doctors come into the room and talk fast, one after another, and the patient doesn't catch half of it. Now the family can look up exactly what was said, who said it, and when. Twice now, something in her past records surfaced a detail one doctor had noticed that another hadn't — and it changed what they looked at.
That's the actual promise here. Not a smarter chatbot. The whole picture, in one place, answerable.