ChatGPT is brilliant for general questions. NotebookLM is better at something narrower and, for research, far more useful: staying grounded in the exact documents you give it.

Most people using AI for research make the same mistake. They paste a chunk of a PDF into ChatGPT, ask a question, and hope for the best — no memory of the source, no way to check what it left out, and no way to search across ten documents at once. NotebookLM, Google's free research tool, was built to solve exactly that problem. It's the deep, grounded end of my research setup; for the wide, live-web end I lean on a cited answer engine instead, which is exactly how I now use Perplexity AI in place of Google for most everyday lookups.

Here's how to set one up properly and turn it into a working research assistant you'll actually keep using.

What NotebookLM actually does differently

Every other AI chat tool draws on a general pool of training data plus whatever you paste into the chat window. NotebookLM flips that. You upload your own sources — PDFs, Google Docs, web pages, YouTube transcripts, even audio — and the model answers only from what you gave it. Nothing outside your notebook gets mixed in.

That single design choice changes what it's good for. It won't guess. It won't blend in half-remembered facts from the wider internet. And every answer comes with an inline citation pointing back to the exact passage it used, so you can verify it in one click. For research, contract review, studying, or pulling a report together from a stack of sources, that's a different category of usefulness than a general chatbot.

Step 1: Create your first notebook

Go to notebooklm.google.com and sign in with a Google account. Click New Notebook. Give it a specific name rather than something vague — "Q3 Competitor Research" beats "Notebook 1" the moment you have more than two of them.

Treat each notebook as a self-contained project, not a junk drawer. A notebook per project, topic, or client keeps the source material focused, which directly improves answer quality — the less irrelevant material in a notebook, the sharper its responses.

Step 2: Add your sources

This is the step that actually matters most. Click Add source and bring in whatever's relevant:

  • PDFs and Google Docs — reports, whitepapers, contracts, research papers
  • Website URLs — paste a link and NotebookLM pulls the page content in directly
  • YouTube links — it reads the transcript, so you can ask questions about a lecture or interview without watching it back
  • Pasted text — notes, emails, anything you've already written
  • Audio files — recordings of meetings or interviews

A single notebook holds up to 50 sources, and each can run up to around 500,000 words, so you can genuinely load an entire project's worth of material into one place. Don't overthink curation early on — add generously, then prune sources later if they're not pulling their weight.

Step 3: Ask questions that stay grounded in your sources

Once your sources are in, the chat panel on the right is where the real work happens. Ask direct questions the way you would of a very well-read colleague who has read only the documents you gave them:

  • "What are the three main arguments across these sources, and where do they disagree?"
  • "Summarise the payment terms across all the contracts I've uploaded."
  • "What did source 3 say about pricing that source 1 didn't mention?"

Every response includes numbered citations. Click one and NotebookLM jumps straight to the exact paragraph it pulled from — this is the feature that makes it trustworthy enough to actually cite in your own work, rather than something you have to double-check from scratch.

"The value of an AI research tool isn't how confidently it answers. It's how easily you can verify it's right."

Step 4: Generate structured outputs, not just chat

Beyond the chat window, NotebookLM can turn your sources into ready-made formats with one click from the Studio panel:

  • Briefing docs — a structured overview of everything in the notebook, useful before a meeting or when onboarding someone else to a project
  • FAQ — auto-generated questions and answers drawn from your material, handy for building internal documentation
  • Study guide — key terms, summaries, and quiz-style questions, built for exam prep or fast onboarding
  • Timeline — chronological breakdown, useful for case research or project histories

These aren't generic templates — they're generated specifically from your uploaded sources, with the same citation trail as the chat answers.

Step 5: Try Audio Overview

This is the feature that gets NotebookLM the most attention, and it's worth trying at least once: click Audio Overview and it generates a podcast-style conversation between two AI hosts, discussing your uploaded material out loud. It's genuinely well produced — natural pacing, back-and-forth banter, not a flat text-to-speech reading.

It's not a replacement for reading closely, but it's an excellent way to absorb a dense report on a commute or a walk, or to sanity-check whether your sources actually say what you think they say before you dig in properly.

Where this earns its keep: loading every past client brief into one notebook before a new pitch, dropping in a stack of research papers before writing a literature review, or uploading a long contract plus your usual playbook so you can ask "where does this deviate from our standard terms?" and get a sourced answer in seconds.

Step 6: Keep your notebooks organised as they grow

A few habits that keep this useful long-term rather than turning into another cluttered tool:

  1. One notebook per active project. Archive or rename notebooks once a project wraps rather than letting sources pile up indefinitely.
  2. Remove outdated sources. If a document gets superseded by a new version, delete the old one — stale sources quietly degrade answer quality.
  3. Use the source guide. The panel on the left lets you toggle individual sources on or off for a given question, so you can narrow focus without deleting anything.
  4. Share notebooks with collaborators. A shared notebook means your whole team is querying the same grounded source set instead of five people each pasting the same PDF into five different chat windows.

Where it falls short

NotebookLM is not a general-purpose assistant. Ask it something outside your uploaded sources and it will tell you it doesn't know, rather than guessing — which is exactly the point, but it does mean you still need ChatGPT or Claude open in another tab for open-ended brainstorming, drafting, or anything that isn't grounded in a specific document set. Think of it as a specialist tool that sits alongside your general AI assistant, not a replacement for one.

The bottom line

The gap between "I asked an AI about this" and "I can prove exactly where this answer came from" is the whole reason NotebookLM exists. For anyone doing real research — students, analysts, anyone reviewing contracts or reports for a living — that grounding is worth more than a chatbot with a bigger general knowledge base.

Set up one notebook today around whatever you're currently working on. Load in five or six real sources. Ask it the question you'd normally spend twenty minutes digging for manually. That's the moment it clicks.

Want more guides like this one? I send a new practical AI workflow to my newsletter every week — sign up here and I'll send the next one straight to your inbox.

Questions people actually ask

Is NotebookLM free?

Yes — NotebookLM is free to use with any Google account at notebooklm.google.com. A single notebook holds up to 50 sources of around 500,000 words each, which is genuinely enough to load an entire project's worth of material without paying anything.

What is NotebookLM used for?

It answers questions using only the sources you upload — PDFs, Google Docs, web pages, YouTube transcripts, pasted text and audio — with inline citations pointing to the exact passage each answer came from. That grounding makes it ideal for research, contract review, studying and pulling reports together from a stack of documents.

Is NotebookLM safe to use with my documents?

Google states that the sources you upload aren't used to train its AI models, and answers are drawn only from your own material. That said, the usual judgement applies to any cloud tool — check your organisation's policy before uploading genuinely sensitive or confidential documents.

What's the difference between NotebookLM and ChatGPT?

ChatGPT draws on its general training data plus whatever you paste into the chat; NotebookLM answers only from the documents you've uploaded, and says it doesn't know rather than guessing. Use NotebookLM when an answer has to be grounded in specific sources, and keep ChatGPT or Claude for open-ended brainstorming and drafting.