⚡ Quick verdict

We may earn a commission if you buy through our links, at no extra cost to you.

Most AI tools will happily answer any question you ask, and that is exactly the problem. Ask a general chatbot about a contract, a research paper or a lecture you missed, and it will answer with confidence — sometimes from its training data, sometimes from nothing at all. The hallucination problem is not a footnote in AI research; it is the reason serious readers still check every claim a chatbot makes, which defeats the purpose of asking in the first place.

NotebookLM was Google's answer to that problem, and it is a genuinely different design. Instead of answering from the whole internet or from its training data, it answers only from the documents you give it. You create a notebook for a subject — a course, a legal case, a product launch, a research topic — fill it with sources, and the AI reads those sources together and answers with citations that point to the exact passage behind each claim. If the answer is not in your sources, the tool is supposed to say so rather than invent one. That single constraint changes everything about how trustworthy the output feels.

It started in 2023 as Project Tailwind, an experiment shown at Google I/O, and it went properly viral in the fall of 2024 when Google added Audio Overviews — two AI hosts discussing your documents like a podcast. Since then it has grown into a full research workspace: Video Overviews, Mind Maps, flashcards, quizzes, slide decks, infographics and reports, all generated from the same sources. In July 2026 Google renamed it Gemini Notebook, folding it into the Gemini brand, and in September 2026 it changed how usage limits work, moving from fixed daily quotas to compute-based limits. The product is the same standalone app at notebook.google — the old notebooklm.google.com address redirects there — but the name, the plans and the limits have all shifted in the last few months, which is why this review dates every fact to 17 November 2026 and tells you where the numbers are still moving.

This review is based on Google's own plans and help pages, the July 2026 rename announcement, and independent guides checked against those pages in November 2026 — not on a paid account or a lab test. Where a figure comes from a third-party guide rather than Google's own page, we say so. For the wider research category, our best AI research tools roundup compares NotebookLM with the field, and our best AI note-taking apps guide covers the note apps it sits next to.

What is NotebookLM — and why is it now called Gemini Notebook?

NotebookLM is a notebook where you drop your materials and Google's AI reads them and helps you work with them. That sentence, simple as it sounds, is the whole product. A notebook is a container for one subject or project, and inside it you collect sources: PDFs, Google Docs, Word files, web pages, YouTube videos, audio recordings, or text you paste in directly. The AI — Gemini models, running with a large context window — reads all of those sources together, so a question can span ten papers at once and the answer can draw on all of them, with each claim tied to the passage it came from.

The rename is the first thing to clear up, because it confuses almost everyone who searches for the tool. On 16 July 2026, Google announced that NotebookLM is now called Gemini Notebook, describing it as the same standalone product with a new name and logo, tied more closely to the rest of the Gemini family. Nothing was shut down: notebooks, files and shared links kept working, and the service gained a second address, notebook.google, alongside the old one. The help pages now call it Gemini Notebook throughout. Most people still search for NotebookLM, which is why this review uses both names — and why you should not be surprised when the app you open says Gemini Notebook on the screen.

The second thing to understand is what source-grounding actually buys you. A general chatbot answers from its training data, which is vast, stale in places, and invisible to you. NotebookLM answers from a set of documents you chose, which you can inspect, add to and remove. That makes the answers checkable in a way a chatbot's answers are not: click a citation and you see the sentence it came from. It also makes the failure mode honest — when your sources do not contain the answer, the tool's job is to tell you that, not to fill the gap with a confident guess. In practice it still sometimes overstates what a source says, so citations are a safety net, not a guarantee — but it is a far stronger safety net than a chatbot gives you.

The third thing is the audience. NotebookLM was built for people who work through material rather than people who want a chat partner: students revising for exams, researchers doing literature reviews, analysts reading filings and reports, lawyers working through case files, product teams reading user interviews, journalists working through documents. If your work starts with a pile of reading, NotebookLM is built for you. If your work starts with a blank page, it is not — the tool has nothing to ground itself in until you give it sources.

How source-grounded answers work in practice

The workflow is short, and that is part of the appeal. Create a notebook, add sources, and start asking questions. Adding a source is a drag-and-drop or a paste: upload a PDF or audio file, paste a web link, import a Google Doc or Slide, add a YouTube video, or paste text directly. Google's source documentation caps a single source at 500,000 words or a 200 MB upload, with separate restrictions for formats such as Slides and Sheets — a limit so high that most real sources never come near it. A single textbook chapter, a 200-page report, a two-hour lecture recording: all of it fits in one slot.

Once the sources are in, the chat is the heart of the product. You ask a question in plain language — "What do these three papers disagree about?", "Summarise the risks section of this filing", "What did the lecturer say about pricing in week four?" — and the answer comes back with numbered citations. Click a citation and the source opens at the relevant passage, so you can check the claim in seconds. That click is the feature that matters most: it turns verification from a chore into a reflex, and it is the reason researchers trust NotebookLM more than a general chatbot for document work.

There are limits to the grounding, and an honest review has to name them. First, NotebookLM can only see what you gave it. It does not browse the open web for fresh information unless you add the page as a source, so a question about something outside your sources gets either a refusal or an answer hedged with caveats — which is correct behaviour, but it means the tool is useless for open-ended research until you have built the source set. Second, YouTube sources have their own rules: only public videos with speech and captions, even auto-generated ones, work, and a video uploaded less than three days ago may not go through yet. Third, the grounding is only as good as the sources: feed it ten biased documents and you get a well-cited biased answer. The tool removes hallucination from the answer; it does not remove bias from the input.

One more practical detail: when you first add sources, some users see a report, flashcards, an infographic, a slide deck or an audio or video overview generated automatically. Google says these are generated once and do not count towards your limits, so the first open of a notebook often feels more generous than the daily quotas suggest. It is a small thing, but it is the difference between a tool that asks you to do all the work and one that shows you what it can do before you ask.

Audio Overviews: the feature that made NotebookLM famous

Audio Overviews are the reason NotebookLM went viral, and in November 2026 they are still the best reason to open the app. One click turns your sources into a podcast-style conversation between two AI hosts who discuss your material — covering the main arguments, pointing out where authors disagree, and explaining technical material in plain language. The voices sound unsettlingly human, the pacing feels like a real show rather than a read-aloud, and the result is genuinely useful: a dense 60-page report becomes a twenty-minute listen you can absorb on a commute or a walk.

There are four formats, and picking the right one matters. Deep Dive is the default: two hosts connecting ideas across your sources, the long-form conversation most people mean when they say "Audio Overview". The Brief is a single speaker delivering the key points in under two minutes — the one to use when you just need the shape of a document before a meeting. The Critique gives constructive feedback on a draft, such as an essay or a design document, with two hosts discussing what works and what does not. The Debate puts two hosts on opposite sides of a question and lets them argue it out, which is surprisingly good for seeing the weak points in your own position before someone else finds them. You can choose a length, pick from more than 80 languages, and add your own instructions to steer the episode — focus on one chapter, skip the background material, spend more time on the disagreement.

Interactive mode is the part most reviews skip, and it is the part that turns a generated file into a conversation. While an episode plays, you can interrupt and ask the hosts a question — "wait, what did the second paper actually conclude?" — and they answer in character before carrying on. It is a small interaction, but it closes the gap between listening and studying: you do not have to go back to the chat to check something you half-heard.

The limits are where the September 2026 change bites. Before 2 September 2026, the free plan was simple: 50 chat questions a day and 3 Audio Overviews a day. Since then, Google has used compute-based limits — heavier requests cost more of your quota, your quota refreshes every five hours until you reach a weekly limit, and the limits may change. November 2026 guides still report 3 Audio Overviews a day on the free tier and 3 Video Overviews a day, with paid tiers multiplying the allowance (Plus twice the standard access, Pro four times), but those daily numbers are now reported figures rather than published guarantees. Treat them as today's best estimate and check the official plans page for the current number before you plan a week's listening around them.

There is also a practical quality note. Audio Overviews are at their best with material that has a real argument in it — papers that disagree, a report with findings, a lecture with a narrative. Feed them a spreadsheet, a terms-of-service document or a set of meeting notes with no through-line, and the hosts will still produce a listenable episode, but it will be padded and vague in the places where the source had nothing to say. The hosts cannot invent substance your sources do not contain, which is the point of the tool — but it means the output is only as interesting as the input.

Video Overviews and the Studio panel

Video Overviews turn the same material into a narrated explainer video — slides and visuals with a voiceover, built from your sources. There are three formats: Explainer, a structured walk-through of the topic; Short, a roughly 60-second version for a quick share; and Cinematic, richer visuals and narrative-led storytelling built with Google's Veo 3 model, which is English-only, for users 18 and over, and exclusive to the Ultra plans, according to November 2026 guides. They are slower to generate than audio, and they land best when you need to share an explanation with someone who will not read the source material — a colleague, a class, a client.

Video Overviews are available in more than 80 languages, though some visual styles may not be available in every language yet, and the same compute-based limits apply: November 2026 reports put the free tier at 3 Video Overviews a day. If you can defer an output for later — Google lets you queue Video Overviews and slide decks rather than generating them immediately — do so when the quota is tight; a deferred generation still counts, but it lets you keep working while the render runs.

The Studio panel is where all of this lives, and it is worth understanding as a whole rather than as a list of buttons. From one panel, your sources become Audio Overviews, Video Overviews, Mind Maps, Reports, Flashcards, Quizzes, Slide Decks and Infographics — every Studio format is included on the free plan, which is the most generous thing about NotebookLM's pricing. The study formats are the ones students actually live in: flashcards and quizzes generated from your lecture notes, a study guide that summarises a whole module, a mind map that shows how the topics connect. The presentation formats — slide decks and infographics — are rougher, and honest testing shows they need editing before you show them to anyone. They are a first draft, not a finished deck.

Two 2026 additions are worth knowing about. First, the September 2026 update added real-time voice conversations with your notebook in nearly 100 languages, according to November 2026 guides — you talk to the notebook and it answers from your sources, which is the closest the product comes to a tutor. Second, the July 2026 rename came with a deeper change: every notebook now has a secure cloud computer, so NotebookLM can write and execute code natively to analyse your sources — sandboxed analysis over your documents, not an agent working in your files. That is a power-user feature most readers will never touch, but it is the direction the product is heading: from a reader of documents to an analyst of them.

Mind Maps: the fastest way to survey a new topic

Mind Maps are the quietest feature in NotebookLM and, for research, one of the most useful. One click produces a branching diagram of the concepts across all your sources — topics, subtopics, and how they connect. For a subject you do not know yet, this is the fastest way to see its shape: generate the map, spot the branch you do not understand, click it, and ask the chat about that topic. For complex material — compliance manuals, technical specs, legal frameworks, multi-author research collections — a Mind Map shows the structure before you read a page, which is the difference between reading with a map and reading blind.

Clicking a node opens an explanation in the chat, so the map is not a static picture but a navigation tool for the notebook. The limitation is platform: Mind Maps work on the web, and the mobile app does not support them yet, according to November 2026 guides — so if you do most of your reading on a phone, the map is a desktop feature for now. The maps are also only as clean as the source material: a well-structured set of documents produces a clean map, while a messy pile of unrelated PDFs produces a messy map with branches that do not quite connect. That is, again, the honest failure mode of a grounded tool — it shows you what your sources actually contain, including their mess.

The notebooks and sources model: what the limits really mean

The notebook is the unit of organisation, and the limits are per notebook and per user, so it is worth understanding exactly what you get. On the free plan you get 100 notebooks, each holding up to 50 sources. That is 5,000 source slots in total — far more than any single person needs for ordinary study or research, and the reason most users will never pay. The binding limit is not the number of notebooks but the sources per notebook: a literature review with 80 papers does not fit in a free notebook, and neither does a legal case with 60 exhibits. That is the moment the paid tiers start to make sense.

Each source can hold up to 500,000 words or a 200 MB upload, with separate restrictions for formats such as Slides and Sheets. One imported item occupies one source slot, whether it is a PDF, a web page, a YouTube video or a pasted block of text, and sharing a notebook does not change the source limit for anyone you share it with. The sources you can add cover most real work: PDFs, Google Docs, Word files, web pages, YouTube videos, audio recordings, and pasted text. Google Docs and other Drive files auto-sync, so a document you edit in Drive updates in the notebook — useful for a living project, and a reason to be careful about which Drive files you connect.

The usage limits are the part that changed most in 2026, and the part where this review has to hedge. Google's plans page now lists the notebook and source counts as the stable, countable limits, and describes chat and generation allowances in relative terms: Plus gets twice the standard access, Pro four times, and Ultra five or twenty times Pro's allowance, depending on the subscription. The old fixed daily counts — 50 chats a day, 3 Audio Overviews a day, 3 Video Overviews a day, 10 daily generations of reports, flashcards, quizzes and mind maps — are still reported by November 2026 guides, but they are reports of the old system, not promises under the new one. Under compute-based limits, a long Audio Overview costs more than a short chat, and your quota refreshes every five hours until you reach a weekly limit. The practical advice is simple: if you hit a limit, wait a few hours rather than assuming the day is lost, and if you hit limits every week, the paid tier is the fix.

NotebookLM limits at a glance — November 2026

Free (Standard)
$0 — 100 notebooks, 50 sources each, standard usage
Google AI Plus
$4.99/month — 200 notebooks, 100 sources each, 2× usage
Google AI Pro
$19.99/month — 500 notebooks, 300 sources each, 4× usage
Google AI Ultra
From $99.99/month — 500 notebooks, 500–600 sources each, 5×–20× Pro usage
Per source
Up to 500,000 words or a 200 MB upload, every tier

Limits are Google's published figures as reported in November 2026 and described as subject to change — check the official plans page before you pay.

Sharing and collaboration

Notebooks are private by default, and sharing is where NotebookLM turns from a personal study tool into a team one. You can share a notebook with specific people, or with anyone who has the link — in the latter case, anyone with a Google account who has the link can open the notebook and ask it questions. A notebook can also be made public, and its owner can then share artifacts such as reports, slide decks or Audio Overviews; anyone with the link to a shared artifact can view it, even if they are not signed in to Google.

Permissions matter, and Google's documentation is blunt about them. A viewer has read-only access to all the source documents and notes you shared with them; an editor can add or remove sources and notes, and share the notebook further with other users. Google also cautions that Chat View — the restricted view that hides the sources — does not completely revoke a viewer's underlying access to the notebook contents, so do not rely on it as a security boundary. The lock icon next to Share means the notebook is private; a shared icon means it is shared with people; a globe means it is public. To make a public notebook private again, set access back to Restricted — new visitors to old links cannot open it.

Advanced sharing is one of the things the paid tiers add: the free plan covers the basics, while upgraded plans unlock more sharing options, according to November 2026 pricing guides. For a study group or a small team, the free sharing is usually enough — a shared notebook where everyone can ask questions of the same source set is genuinely useful, and it is the closest NotebookLM comes to a collaborative research tool. For a classroom or a company, the Workspace routes add admin controls, audit logs and data-loss prevention on Enterprise, Frontline and Education editions, plus external sharing that has been off by default since 10 September 2026 unless an admin allows it.

The sharing model has one sharp edge worth stating plainly. When you share a notebook, you are sharing the sources, not just the answers. A viewer can read every document you put in, and an editor can change the set. That is the design — a shared notebook is a shared library — but it means you should never put a document in a notebook you might share unless you are comfortable with every viewer reading it. Keep separate notebooks for private material and shared material, and check the Share icon before you send a link.

Pricing: free is the real product, paid is headroom

NotebookLM is free, and that is not a trial or a teaser — the free plan includes every Studio format, 100 notebooks and 50 sources per notebook, which is enough for almost any student or solo researcher. Paid access comes through Google AI plans rather than a separate NotebookLM subscription: there is no "NotebookLM Plus" to buy any more, and any price table still selling it under that name belongs to an earlier naming scheme. The current personal tiers, from Google's US subscriptions page as reported in November 2026, are Google AI Plus at $4.99 a month, Google AI Pro at $19.99 a month (or $199.99 a year), and Google AI Ultra at $99.99 or $199.99 a month. Each is a Google AI bundle that also includes storage and the Gemini app — Plus comes with 400 GB, Pro with 5 TB, Ultra with 20 TB or 30 TB — so the price buys more than NotebookLM limits.

What each tier actually changes is headroom. Plus doubles the free plan's notebooks to 200 and its sources to 100 per notebook, with twice the standard usage allowance. Pro raises that to 500 notebooks and 300 sources per notebook, with four times the standard usage. Ultra keeps the 500-notebook ceiling and raises sources to 500 or 600 per notebook, with five or twenty times Pro's usage depending on the tier, plus the highest model access, Cinematic Video Overviews, and co-created slide decks and Video Overviews. There is also a work route: a NotebookLM licence on Google Cloud is reported at $9 per licence per month after a 30-day trial, with a minimum of 15 licences, for organisations that want NotebookLM by itself rather than the whole Google AI bundle.

The honest pricing advice is to stay free until a specific limit bites. If no notebook needs more than 50 sources and your generation schedule works within the compute-based quota, Plus buys you nothing you need — $4.99 a month for headroom you do not use is still $4.99 a month. The upgrade that earns its money is Pro at $19.99: 300 sources per notebook covers a real literature review, four times the usage covers a heavy week, and the 5 TB of storage and the Gemini app access make it a bundle rather than a single-tool purchase. Ultra at $99.99 is for people who live in the tool all day — researchers, analysts and teams generating video and decks constantly — and at that price you should be sure the usage is there before you commit. Google's own pages describe the limits as subject to change, and the plan names have already changed twice in two years, so re-read the plans page before you pay rather than budgeting on this review's numbers.

One more pricing note: Google applies a visible watermark automatically if you live in India, South Korea or Vietnam, and watermark removal is listed as a feature of the upgraded plans rather than the free one, according to November 2026 guides. If you share Audio or Video Overviews publicly from one of those countries, that watermark is part of the output on the free plan.

Privacy: what happens to your documents

Privacy is the question every serious user asks before uploading a contract, a thesis draft or a set of interview transcripts, and Google's stated policy is clearer than most AI tools — with one catch that matters. Google states that the content in NotebookLM will not be used to directly train its foundational AI models, unless you choose to provide feedback. That is the headline, and it is a good one: your sources and chats are not training data by default.

The catch is feedback. When you send feedback — a thumbs up or down, or a problem report — Google may review the full context of that interaction, including your queries, uploads and the model's responses. That feedback is reviewed by specially trained teams, disconnected from your Google account before reviewers see it, and retained for up to three years. Google's own guidance is blunt: do not include confidential or sensitive information in your feedback. In practice, that means the thumbs-up button is a data-sharing decision, not a casual gesture — do not tap it on a notebook full of client material.

Workspace and Education accounts get stronger protection. Google says that for Google Workspace or Workspace for Education users, uploads, questions and answers will not be reviewed by human reviewers even when you provide feedback, and will not be used to train AI models — with the work-account table limiting that handling to editions where NotebookLM is a core service or add-on. There are edge cases to know about: data that NotebookLM shares with other Google services, such as the Gemini app, is used under the Google Privacy Policy and the relevant service-specific notices, including for product improvement, and chats with your notebook in AI Mode may be used to improve Google's generative AI models. Google does not publish a NotebookLM-specific retention period; its general retention policy says data you can delete stays in your account until you remove it, with deletion generally taking about two months and backups kept up to six months.

The practical framework is simple. NotebookLM is private — your sources are not publicly visible by default — but it is not a vault for confidential data. Do not upload anything you could not afford to have reviewed if you accidentally send feedback, and do not upload client or company material on a personal account without checking your firm's rules, because a file uploaded to NotebookLM becomes a copy that follows the product's sharing settings and retention, not the original file's permissions. On a personal account, no admin console can see or delete that copy, and it leaves with the employee. For genuinely sensitive work, use a Workspace account on a qualifying edition, where the no-review and no-training guarantees apply.

Who NotebookLM is best for — and who should skip it

NotebookLM is a good fit if: you are a student working through lecture notes, readings and past papers — the flashcards, quizzes, study guides and Audio Overviews are the best free study stack we have seen; a researcher doing a literature review across dozens of papers, where the citation trail is the whole point; an analyst reading filings, reports and transcripts, where a grounded summary beats a confident guess; a lawyer or paralegal working through case files, where every answer needs a passage behind it; or a product team synthesising user interviews, where the sources are the evidence and the AI is the analyst.

NotebookLM is not a good fit if: you want a general chatbot — it cannot answer from anything outside your sources, so it is useless for open-ended questions until you have built the source set; you need real-time information, since it does not browse the open web on its own; you work mostly on a phone, where Mind Maps are not supported yet and the small screen fights the notebook layout; or you handle confidential client material on a personal account, where the feedback and retention rules above make it the wrong home for that data.

There is also a workflow nuance worth stating. NotebookLM rewards people who organise their material before they ask questions: a notebook per subject, clean sources, a clear question. People who open it with a vague question and no sources will get a vague answer and conclude the tool is weak — when the weakness was the input. The tool is a research assistant, not a mind reader, and the ten minutes you spend curating sources is what makes the next ten hours of answers trustworthy.

NotebookLM (Gemini Notebook)

Strengths

Weaknesses

NotebookLM vs the rivals

NotebookLM is not alone in the research category, and the comparison decides whether it is the right tool or just the most famous one. ChatGPT is the all-rounder: it writes, codes, browses the open web, remembers across conversations, and its Projects feature holds documents too — but its answers are grounded in training data first and your files second, and citations are not the default. For open-ended work, ChatGPT wins; for document-grounded work, NotebookLM wins. Our ChatGPT vs Gemini comparison covers the all-rounder head-to-head, and Gemini itself is the closest sibling — the same models, the same Google account — but Gemini answers from the open web and its training data, not from a closed source set you control.

Perplexity is the other serious rival, and the contrast is instructive. Perplexity is a research engine for the open web: it searches live, cites its sources, and is superb for finding out what is true right now. NotebookLM is a research engine for your documents: it never leaves the source set, and it is superb for finding out what your material says. They are complements, not substitutes — Perplexity to discover, NotebookLM to digest. Our Perplexity vs ChatGPT comparison and our best AI research tools roundup put the whole field side by side, including the tools that specialise in academic papers and the ones that build on top of your notes.

Notion AI and the note-taking apps are a different kind of rival. Notion AI works inside your notes and can summarise and rewrite them, but it is a writing assistant with AI features, not a research workspace with citations — and it costs money from the start. NotebookLM is free and citation-first. If your notes already live in Notion, our best AI note-taking apps guide compares the options; if your work starts with a pile of documents rather than a note database, NotebookLM is the better home for it. And for researchers turning reading into published content, our best AI SEO tools guide covers the tools that take research and turn it into articles that rank — NotebookLM does the reading, an SEO tool does the publishing.

What to watch before you trust it with real work

Four things are worth watching as NotebookLM matures. First, the compute-based limits. The September 2026 change from fixed daily quotas to a compute-based system is the biggest usability change in the product's history, and nobody outside Google knows exactly how a long Audio Overview is priced against a short chat. The first month's quota behaviour will decide whether heavy users need Plus or Pro, and the numbers are not public yet — watch your own usage before you pay.

Second, the rename is still rippling through documentation. Google's help pages now say Gemini Notebook, third-party guides still say NotebookLM, and plan pages mix the two names. That is normal for a rename, but it means you should verify any price or limit on Google's own plans page rather than taking a single guide — including this one — as final. The plan names have changed twice in two years, and they may change again.

Third, the agentic direction. The July 2026 addition of a secure cloud computer per notebook — letting NotebookLM write and execute code to analyse your sources — is the start of a shift from reader to analyst. It is powerful for data work and it is also a new surface for mistakes: code that analyses your sources can also misread them, and the output needs the same citation-checking reflex as the chat. Treat agentic outputs as a draft, not a result.

Fourth, the mobile gap. NotebookLM has mobile apps, but Mind Maps do not work on them yet, and the notebook layout is a desktop-first design. If your reading happens on a phone — on a commute, in a waiting room — the experience is thinner than the desktop one, and the Audio Overviews, which are the most mobile-friendly feature, are the part that carries the app. Watch for the mobile feature gap to close; until it does, plan your deep work for a desk.

Our verdict

NotebookLM is the most trustworthy AI research tool most people will ever use for free, because it solves the one problem that makes every other chatbot hard to trust: it shows its work. The source-grounded answers with clickable citations, the generous free plan, and the Audio Overviews that turn a reading list into a podcast are a combination no rival matches at $0. For a student, a researcher, or anyone whose work starts with a pile of documents, it is the first tool to open — before the chatbot, before the search engine, before the highlighter.

But it is a specialist, not a generalist, and the verdict has to say so. It cannot answer from outside your sources, its limits are now compute-based and hard to predict, and its presentation formats need editing before they are presentable. The right way to use it is as one tool in a stack: NotebookLM for grounded research, a general chatbot for everything else, and a dedicated SEO or writing tool for turning the research into published work. Used that way, it earns its place in a daily workflow within a week — and it costs nothing to find out whether it earns its place in yours.

Bottom line: the best free research assistant Google makes, and the only mainstream AI tool that cites every claim. Start free, curate your sources carefully, and check the official plans page before you pay for headroom you may not need.

Try NotebookLM →

Frequently asked questions

Is NotebookLM really free?

Yes. NotebookLM — now called Gemini Notebook — is free with any Google account: 100 notebooks, up to 50 sources per notebook, and every Studio format, including Audio Overviews, Video Overviews, Mind Maps, flashcards and quizzes, as checked on 17 November 2026. Paid Google AI plans raise the limits: Plus at $4.99 a month doubles them, Pro at $19.99 a month quadruples them, and Ultra from $99.99 a month raises them further. Google's limits are described as subject to change, so check the official plans page before you pay.

What is NotebookLM called now?

Google renamed NotebookLM to Gemini Notebook on 16 July 2026 and calls it the same standalone product. The old notebooklm.google.com address redirects to the new notebook.google address, and your notebooks, files and shared links kept working. Most people still search for NotebookLM, which is why this review uses both names.

How many sources can NotebookLM handle?

On the free plan, a notebook holds up to 50 sources, and each source can hold up to 500,000 words or a 200 MB upload, as checked on 17 November 2026. Google AI Plus raises that to 100 sources per notebook, Pro to 300, and Ultra to 500 or 600 depending on the tier. One imported item occupies one source slot, whether it is a PDF, a web page, a YouTube video or a pasted block of text.

What are Audio Overviews?

Audio Overviews turn your sources into a podcast-style conversation between two AI hosts who discuss your material, point out where authors disagree, and explain technical ideas in plain language. There are four formats — Deep Dive, the default two-host conversation; the Brief, a single speaker in under two minutes; the Critique, feedback on a draft; and the Debate, two hosts arguing different sides — and an interactive mode lets you interrupt and ask the hosts questions mid-episode. Free users get 3 Audio Overviews a day according to November 2026 reports, though Google has moved to compute-based limits, so check the official page for the current number.

Does NotebookLM train on my documents?

Google says the content in NotebookLM will not be used to directly train its foundational AI models, unless you choose to provide feedback. If you send feedback — a thumbs up or down, or a problem report — Google may review the full context of that interaction, including your queries, uploads and the model's responses, and that feedback can be retained for up to 3 years, disconnected from your account. Workspace and Education accounts get stronger protection: uploads, questions and answers are not human-reviewed and are not used to train models. Do not upload confidential material you would not want reviewed.

Is NotebookLM better than ChatGPT for research?

For research grounded in your own documents, yes — NotebookLM answers only from the sources you give it and shows a clickable citation for every claim, which makes it much harder for the model to invent facts. ChatGPT is the better all-rounder: it writes, codes, browses the open web and remembers across conversations, while NotebookLM cannot answer from anything outside your sources. The honest stack is both: NotebookLM for source-grounded research, ChatGPT or Gemini for everything else.

AT
About Aitopkit

We research AI tools in depth before recommending them — official pricing, public docs, and real user reports. If you buy through our links we may earn a commission at no extra cost to you. Read our disclosure.