Built by Google, the tool is an AI research and note-taking assistant. Upload PDFs, documents, slides, web pages, or YouTube links and the assistant turns them into summaries, audio overviews, study guides, and an interactive chat that cites every claim back to the source it came from.
Every answer is anchored to a passage in the material you uploaded, with inline citations you can click through to verify.
Turn a stack of sources into a two-host audio overview you can listen to on the commute, the treadmill, or a dog walk.
The tool runs on Google's long-context Gemini models, so it reads whole books and lengthy reports — not just a few paragraphs.
NotebookLM coverage appears in
A workspace where every answer carries a footnote back to the source it came from.
NotebookLM sits somewhere between a research assistant and a note-taking app. You create a notebook, drop in up to a few hundred sources, and the tool reads the whole collection with a long-context Gemini model. From there, it can write briefing docs, build study guides, turn a reading list into a twelve-minute conversation between two AI hosts, or answer questions as if it were a teaching assistant who has memorised every file you uploaded.
The thing that makes this research assistant different is the grounding rule. It is not trying to be smart about the world at large — it is trying to be smart about your sources. When the assistant writes an answer, every paragraph links back to the passage it drew from. If two sources disagree, the notebook points out the disagreement rather than papering over it. That single design choice is what turns NotebookLM from a novelty into something a librarian, a paralegal, or a medical student can actually trust to sit next to their work.
Under the covers the tool is built around three loops: an indexing pass that chunks and embeds each source, a retrieval pass that pulls the most-relevant passages for a given prompt, and a generation pass where Gemini composes an answer and links every sentence to its citation. All three loops run on Google infrastructure; the indexing loop never leaves the account boundary you started from.
Press one button and the tool produces a conversational podcast about your material.
In September 2024 Google quietly added a button called "Audio Overview" to NotebookLM. You click it, the app reads your sources, and a few minutes later you get a two-host podcast that explains the material as if two knowledgeable colleagues were discussing it over coffee. The result is uncanny enough that students, researchers and product managers started sharing clips of the audio on social media within days of launch.
Behind the scenes the tool is stitching together a long-form dialogue: it picks the through-line, decides which host asks which question, varies the tempo, and mixes in pauses and small-talk that make the output sound less like text-to-speech. You can now customise tone, focus topics, interject while the hosts are speaking, and download the MP3. Audio overviews are available for every notebook you own, including ones with YouTube transcripts as the only source.
Typical overviews run 8 to 18 minutes. The app limits length partly to keep the output tight and partly to manage the compute cost on the free tier; the paid tier raises the per-day cap significantly. Since 2025 the feature has added three further modes — a short "Brief" under four minutes, an extended "Deep Dive" that can run past forty, and a study-oriented "Critique" mode where the two hosts probe the material for weak arguments.
Researchers, lawyers, teachers, and product managers pick the tool because the citations make review possible.
The single biggest reason a team reaches for this assistant instead of a generic chatbot is auditability. A paralegal cannot submit a brief that cites "an AI answer." A medical student cannot revise from a summary that may have hallucinated a number. The notebook solves that friction by making every sentence traceable: click the footnote and you are taken to the exact paragraph on page 14 that the model drew from.
Educators have been the loudest adopters. A notebook full of a semester's reading list becomes a patient tutor that will re-explain any concept at any level, generate practice questions, or read a chapter aloud as an audio overview. Researchers use the tool the way they used to use a stack of index cards — except the assistant also writes draft paragraphs that cross-reference three papers at once, flags where the argument is thinnest, and proposes further reading based on what is already in the corpus.
The product does not replace your brain. It replaces the forty minutes you used to spend skimming a stack of PDFs to find the one paragraph you half-remembered, and the two hours you used to spend untangling why three authors seemed to be contradicting each other. See the U.S. NIST AI guidance for a policy-level view on how such tools fit into risk-managed workflows.
A notebook can ingest hundreds of files at once — and every answer will cite the slice it used.
A notebook is only as good as the sources inside it. The tool currently accepts PDFs, Google Docs, Google Slides, plain-text files, Markdown, arbitrary web URLs, YouTube videos with transcripts, and raw pasted text. During 2024 and 2025 Google began a staged rollout of audio-file ingestion and spreadsheet support; both are live for Plus-tier subscribers and in partial rollout on the free tier.
The free tier currently caps a notebook at fifty sources and roughly 500,000 words of combined text. The Plus tier raises that to three hundred sources per notebook and a much higher word-count ceiling. Sources can be reordered, renamed, hidden from the model temporarily, or removed entirely without losing your notes. A single source can be cited by every note you save inside the same notebook, which is what makes the tool function like a research carrel rather than a chat window.
Citations in the tool are not decoration. When a reader clicks a footnote, the exact passage — often a single paragraph or even a sentence — is highlighted in the source pane. That behaviour is the feature most quoted by reviewers from the Stanford AI portal and it is the reason many enterprise teams have moved their intake reading into a shared notebook rather than a shared folder.
Longer quotes from people running NotebookLM in real work — swipe to see more.
"NotebookLM cut the time I spend digesting a new corpus by roughly two-thirds. I upload the reading list, generate a 14-minute audio overview for the drive home, and by the time I sit down at my desk the next morning I already know which three papers deserve a second pass."
— Harriet O. Vondracek-LundinResearch Lead, Oakbridge Linguistics Institute, Reykjavík
"We gave every graduate archivist access to NotebookLM for two months. The finding-aids team went from producing one draft inventory a week to three. The grounding citations matter — we never publish a summary without verifying the footnote, but NotebookLM puts you right at the source."
— Cyprian J. AchterbergArchivist, Meridian Heritage Foundation, Bergen
"The first time my Year 12 students heard a NotebookLM audio overview of their set text they laughed, then went quiet, then asked for one for every unit we had left. Attendance at revision sessions rose. I am not a convert to every AI tool, but NotebookLM earns its place."
— Solveig P. Tremblay-RousseauCurriculum Director, Northwind Preparatory Academy, Québec
"I drafted my doctoral literature review in NotebookLM, one chapter at a time. NotebookLM never once let a citation drift. When I asked for a synthesis across seven papers, it produced one — and pointed me at a real disagreement between two authors I had not noticed after four months of reading."
— Ingrid Z. Lindqvist-MorenoDoctoral Candidate, Juniper Hill Graduate School, Utrecht
"Our policy team runs a NotebookLM notebook for each live brief. The notebooks contain hearing transcripts, committee reports, and statutes. The speed is obvious, but the thing that surprised us is how much NotebookLM keeps us honest — vague language in our own briefs gets flagged the moment we try to paraphrase past it."
— Alaric B. WesterveldPolicy Director, Stonebridge Civic Institute, Tallinn
"I asked NotebookLM to produce a revision guide for a unit I had taught six times. It surfaced three connections I had never made explicit in class. Reading NotebookLM's notes felt like being given feedback from a very well-read colleague who had audited every session."
— Desirée K. Okonkwo-HartmannSenior Lecturer, Fernway Institute of Education, Lausanne
Short, grounded answers to the things that come up when someone discovers NotebookLM for the first time. Longer walkthroughs live on the dedicated pages.
NotebookLM is Google's AI-powered research and note-taking assistant. You upload sources — PDFs, Google Docs, slides, URLs, YouTube videos, or pasted text — and NotebookLM uses Gemini models to summarise, cross-reference, and answer questions that are grounded in those sources. Every NotebookLM answer includes citations back to the passages it used.
The standard NotebookLM tier is free for anyone with a Google account and covers most individual use cases. A paid tier called NotebookLM Plus raises the source limit per notebook, increases the number of audio overviews you can generate per day, and adds team sharing and analytics. See the NotebookLM pricing breakdown for the current caps.
No — Google states that personal-account NotebookLM uploads are not used to train its foundation models. Workspace-tier uploads fall under the same Workspace data-handling commitments that cover other Workspace apps. The NotebookLM data and privacy page goes deeper, and the FTC business-guidance hub on privacy has a regulator's take on evaluating AI tools.
NotebookLM runs on whichever Gemini long-context model Google currently considers production-ready. Historically the tool has moved from the PaLM 2 stack behind the original Project Tailwind prototype, through Gemini 1.5 Pro, and on to the Gemini 2.x family during 2025. See the Gemini plus NotebookLM deep-dive for the full lineage.
At the time of writing: PDF, Google Docs, Google Slides, plain text, Markdown, web URLs, YouTube links, and pasted raw text. Audio ingestion and spreadsheet support have been rolling out through 2024 and 2025. The sources and uploads guide tracks the live list.
Yes. Google shipped official NotebookLM apps for Android and iOS in May 2025. The apps add background audio-overview playback, offline listening, and a share-sheet extension that lets you send a web page or PDF straight into a notebook. The NotebookLM app page lists current OS version requirements.
The tool is source-grounded by design — every answer cites the passage in your uploaded material. General chat tools like ChatGPT or Claude are open-domain by default and will happily answer without a source corpus. This research notebook also generates podcast-style audio overviews, which the general chatbots do not natively produce. Neither tool replaces the other; they solve different problems. Plenty of workflows use a general chatbot for brainstorming and the notebook for anything that needs to be traceable to cited material.
Yes. Every notebook has a share sheet with granular roles (viewer, commenter, editor). On the Plus tier, administrators can set notebook-level retention policies and track which members have opened which sources. For classrooms, the tool also supports read-only links that let students listen to an audio overview or run a Q&A session without being able to edit the underlying sources. Sharing does not duplicate your sources — collaborators see the same indexed corpus you do.
Yes. Chat, summaries and note generation have been multilingual since mid-2024. Audio overviews added Japanese, Spanish, Portuguese, German, French, Italian, Korean and Hindi voice hosts through 2025, with further languages in rotation. Source ingestion has always been language-agnostic — you can upload French papers and ask questions in English, and the citation will still resolve to the original passage.
Drop in a paper, a reading list, or a YouTube playlist. NotebookLM turns it into summaries, study guides, and audio overviews in under two minutes.
Walk through the first-notebook flowIf you are landing here for the first time, the most-read pages tend to be the NotebookLM features rundown, the hands-on how-to-use-NotebookLM walkthrough, the long-form NotebookLM guide, and the focused NotebookLM tutorial for first-time users. Anyone curious about the pricing tiers usually starts with the NotebookLM pricing breakdown before looking at the NotebookLM Plus paid tier. People evaluating the tool typically also read our NotebookLM review and the NotebookLM demo walkthrough before committing to a workflow.
Other popular deep-dives include the Gemini and NotebookLM architecture page, the everyday NotebookLM notes studio, the overarching NotebookLM AI primer, the origin-story NotebookLM history, the NotebookLM online access guide, the NotebookLM web client reference, the NotebookLM website walkthrough, the NotebookLM app page for Android and iOS, the disambiguation page for the Google LM notebook name, the LLM notebook concept explainer, the NotebookLM capabilities list, the NotebookLM Google product context, and if you prefer reading source material first, the Stanford AI research portal hosts a strong external overview of long-context language-model tooling.