How to Use NotebookLM in 2026 (It’s Called Gemini Notebook Now)

Transparency note: Some links in this article are affiliate links. If you sign up through them, I may earn a small commission at no extra cost to you. I only recommend tools I actually use and test myself.

Last month I had 41 PDFs sitting in a folder — insurance documents, two contracts, a stack of research papers, and a 90-page onboarding manual I was supposed to have read. Reading all of it properly would have cost me about six hours.

I dropped the whole folder into NotebookLM instead and had my answers in 25 minutes.

Here’s the thing nobody tells you when they explain how to use NotebookLM: it isn’t a chatbot. Ask it to write you a poem and it’ll shrug. It only knows what you feed it — and that restriction is the entire point. This guide walks through the three panels, the Studio tools worth your time, what the free tier actually gives you, and the four mistakes that cost me my first month.

One more thing before we start, because it’s confusing a lot of people right now.

Wait — it’s called Gemini Notebook now

On July 16, 2026, Google renamed NotebookLM to Gemini Notebook. Same product, same team, same URL. Your notebooks didn’t move and nothing broke.

The “LM” stood for language model, which meant nothing to most people, and Google has been folding everything under the Gemini brand for a while now — Bard went first, then Duet AI. This was next in line.

Practically, it changes two things for you. The logo is now a blue-purple gradient, and half the tutorials you’ll find still use the old name. I’ll use both names here so nothing gets lost. You still get there at notebooklm.google, and it’s still free to start with a Google account.

Pro tip: If you search for help and land on a guide that doesn’t mention the rename, check the publish date. Anything written before mid-July 2026 will be missing the Studio tools added this year.

The three panels, left to right

Open a notebook and you get three columns. The trick is to work through them in order, left to right. That’s it — that’s the whole mental model.

• Sources (left) — everything you want it to read. PDFs, Google Docs, Sheets, Slides, web pages, YouTube links, pasted text, audio files.
• Chat (middle) — where you ask questions. Every answer comes with a numbered citation you can click to jump to the exact line in the source.
• Studio (right) — where you turn all that into something you can actually send someone. Reports, slide decks, mind maps, quizzes, audio.

Most people stop at the middle column. That’s the mistake. Chat is useful, but Studio is what makes this different from pasting a PDF into ChatGPT.

Step 1: Add your sources (don’t rush this bit)

Click Add source and upload. You can drag a whole folder in at once. Each source can be up to 500,000 words or 200MB, which is roughly a 1,500-page book — you’re unlikely to hit it.

More importantly, there’s a distinction here that took me weeks to notice and it matters more than anything else on this page. Google Docs, Sheets and Slides behave as living documents. Edit the original, hit refresh on the source, and your notebook updates. PDFs and uploaded files are frozen at the moment you added them.

However, I learned this the annoying way. I had a budget spreadsheet exported as a PDF, updated the real spreadsheet twice over two weeks, and kept getting answers from the stale numbers. Twenty minutes of confusion for no reason. Now anything I expect to change goes in as a Drive file, never an export.

Use live sources when the facts can change

Also, if you don’t have sources yet, the search box at the top of the Sources panel will find them for you. Type what you’re looking for in plain English and it returns a list of web results to pick from — like Google Search, except each result becomes a source with one click.

Pro tip: Give each notebook one job. A notebook with 40 sources across four unrelated topics gives noticeably vaguer answers than four notebooks with 10 focused sources each. Free accounts get 100 notebooks, so there’s no reason to hoard.

Step 2: Configure chat before you ask anything

There’s a Configure chat option above the message box. Almost nobody touches it, and it’s the single biggest quality upgrade available to you.

It sets a standing instruction applied to every answer in that notebook. So instead of re-explaining your situation each time, you say it once. This is what I paste into any notebook holding contracts or policy documents:

You are helping me review documents I have to make a decision on. Answer in plain English at a level a smart 15-year-old would follow. Always quote the exact wording from the source before you interpret it. If the sources disagree with each other, say so explicitly instead of picking one. If something isn’t covered in my sources, say “not in your sources” — never fill the gap.

That last line matters. Gemini Notebook is already far more grounded than a general chatbot, but telling it to flag gaps rather than paper over them is what makes the answers trustworthy enough to act on.

Meanwhile, one more habit worth building: clear your chat history every few exchanges. Long threads drift, and old questions start coloring new answers. Before you clear, save anything useful as a note — and if it’s really useful, convert that note into a source so it feeds every Studio output afterwards.

Pro tip: Ask “what’s in these sources that I probably haven’t thought to ask about?” It’s the question that has surfaced the most new information for me, because it doesn’t depend on me already knowing what to look for.

Step 3: The Studio tools, ranked by how much I use them

Next, Studio has nine tools now. They are not equally good. Here’s my honest split after using this thing most weeks since last year:

Mind Maps are the most underrated thing in the product. Generate one the moment you finish adding sources — before you read a single page. You get the whole shape of the material in about fifteen seconds, and every branch is clickable, which opens a chat grounded in just that topic. It’s how I decide what’s worth reading properly and what I can skip.

Reports is where the real time goes back in your pocket. Skip the default format buttons — they’re generic. Use the Suggested formats underneath, which are generated from your actual sources. Feeding it four competitor pricing pages and taking the suggested comparison brief turned a two-hour job into about eight minutes for me.

Audio Overviews get the most attention online and I’d argue they deserve the least. Two synthetic hosts chat about your documents for ten minutes. It’s a neat trick, and for anything I need to know, typing the question into Chat gets me there in twenty seconds instead. The exception is commuting — I turn long newsletters into audio and listen while walking. That’s a real use, just a narrow one.

Pro tip: Upload your brand guidelines or a style sheet as a source, then add “follow the attached brand guideline for colors, fonts and design style” to your prompt. Infographics, Reports and Slide Decks all respect it.

Free tier reality check: what you actually get

In practice, the free plan is more generous than most people assume, and every core feature is included — Audio Overviews, Video Overviews, Deep Research, Slide Decks, all of it. You’re limited on volume, not on capability.

Fifty chat questions a day sounds tight and mostly isn’t — I’ve hit it maybe twice, both times while stress-testing a contract. The limit you’ll actually feel is 50 sources per notebook. If you’re a student pulling together a dissertation reading list or a lawyer reviewing a document set, you’ll bump into that ceiling fast, and it’s the one honest reason to pay.

Google also sells a cheaper Plus tier that sits between the two with 100 sources per notebook, plus Ultra plans starting at $99.99/month for very heavy use. Check the current pricing page before you commit — Google has moved these numbers more than once this year.

Pro tip: Don’t pay on day one. Run the free tier for two weeks. If you never hit 50 sources in a single notebook, Pro buys you nothing you need. If you’re weighing up subscriptions generally, I keep a running list of AI tools under $10/month that are worth paying for.

Four mistakes I made in my first month

1. I treated it like ChatGPT. I asked it to draft an email from scratch and got a flat refusal, more or less. It won’t invent things that aren’t in your sources. That’s a feature — it’s why the citations mean something — but it means Gemini Notebook is the wrong tool for blank-page work. For that I still open ChatGPT.

2. I dumped everything into one giant notebook. Forty-plus mixed sources, four unrelated projects. Answers got noticeably mushier. Splitting them into focused notebooks fixed it in an afternoon.

3. I ignored the citations. Every claim has a little numbered chip next to it. Click it and you land on the exact sentence in the source. For about six weeks I skimmed past them — which meant I was trusting the summary without ever checking the original. Two clicks is not a lot to ask when the answer is going into something that matters.

4. I expected editable slide decks. More on this next, because it’s the gotcha most guides skip.

What NotebookLM won’t do

Three honest limitations, because you’ll find these out anyway and it’s better to know now.

Slide decks export as flat images. Download a generated deck as PowerPoint and every slide is a picture. You can’t edit the text, move a box, or restyle anything. Google added prompt-based revision in February 2026 — you can say “make slide 5 more concise” and it rebuilds the deck — but you never get editable elements. So use it to generate a narrative, then rebuild the real deck yourself. If you want something you can actually edit, I’ve written up how to use ChatGPT for PowerPoint instead.

The built-in Deep Research is the weakest version of Deep Research. It’ll go find sources on the open web and write them up, which solves the blank-notebook problem. But compared to what Gemini, ChatGPT and Claude produce with their own research modes, the output is thinner. I do my research elsewhere and bring the good sources back in. If you’re comparing options there, my roundup of the best AI tools for research covers the ones I actually use.

It has no memory between notebooks. Every notebook is sealed off. Something you established in one is invisible in another. Useful for privacy, irritating when you’re working across related projects.

Skip it entirely if: you mainly want an assistant to write, brainstorm or code. That’s not what this is. Gemini Notebook only earns its place when you already have the documents and the problem is getting through them. For the writing side of Google’s stack, see my guide to how to use Google Gemini.

Frequently asked questions

Is NotebookLM free?

Yes. You get 100 notebooks, 50 sources per notebook, 50 chat questions a day, and three Audio and three Video Overviews a day at no cost with a Google account. Every feature is available on the free tier — paid plans raise the limits, they don’t unlock tools.

Why is NotebookLM called Gemini Notebook now?

Google renamed it on July 16, 2026 to bring it under the Gemini brand, the same way Bard and Duet AI were folded in earlier. It’s the same product at the same address with the same notebooks — only the name and logo changed.

How do I use NotebookLM to study?

Upload your lecture slides, readings and notes into one notebook per subject. Generate a Mind Map first to see the shape of the material, then use Flashcards for terminology and Quiz to test yourself. Ask the chat to explain anything you got wrong, and it’ll answer from your own materials rather than the general internet. My guide to using ChatGPT for studying pairs well with this.

Can NotebookLM make a podcast from my documents?

That’s the Audio Overview feature — two AI hosts discussing your sources in a podcast format. You can pick Deep Dive, Brief, Critique or Debate, and steer what they focus on with a prompt. Free accounts get three a day. It’s polished, though for most questions typing into Chat is faster.

Does NotebookLM make things up?

Far less than a general chatbot, because it answers only from the sources you’ve given it and cites the exact line for each claim. It can still misread or over-summarize, which is why the citations exist. Click through on anything you’re going to rely on.

Start with one folder

Pick the pile you’ve been avoiding — the insurance renewals, the onboarding pack, the four papers you keep meaning to read. Make a notebook, drop them in, generate a Mind Map, and ask it what you should be worried about.

Ten minutes. That’s the entire commitment, and it’s how I got the six hours back. If it clicks, the next thing to set up is a standing Configure chat instruction — that’s where the answers go from good to worth acting on.

Want more of these? I test one AI tool properly every week over at the best AI productivity tools, ranked by hours saved rather than feature count.