What is NotebookLM (now Gemini Notebook) and how to use it for work

What NotebookLM is, and why it is now called Gemini Notebook

NotebookLM started as a Google Labs experiment built around one idea: a notebook with a language model at its core (the "LM" in the name) that works from material you choose rather than from everything it absorbed in training. Google's July 2023 launch post called this source grounding. A notebook is a collection of sources for one project. The model reads them, summarizes them and answers questions about them, and each answer carries inline citations that point to direct quotes, text or images in your sources.

The difference from a general chatbot is the boundary. Google's help pages say chat responses only use data from your sources, and if the answer is not in the source material, it won't provide a response. That makes it a tool for understanding documents you already trust, not for asking the internet a question.

On July 16, 2026, Google renamed NotebookLM to Gemini Notebook. It remains a standalone product, now at notebook.google.com (old notebooklm.google.com links redirect there), and Google put its reach at more than 30 million people and over 600,000 organizations. Google tied the new name to the product doing more across its ecosystem: notebooks now also open in the Gemini app and in AI Mode in Google Search. The models underneath are Google's Gemini family, whose versions and naming are explained in our post on what Google Gemini is. This post uses NotebookLM for the history and Gemini Notebook for how the product works today.

NotebookLM release history, from Project Tailwind to Gemini Notebook

Every date below comes from Google's own announcements.

Two threads run through that list. The model underneath keeps changing, from Gemini Pro to 1.5 Pro, 2.0 Flash and, since June 2026 on the top plans, Gemini 3.5, so reviews written in 2024 describe a different engine. And the product has moved from a personal experiment to a work tool, with Workspace terms, admin controls and paid tiers.

How NotebookLM works: answers grounded in your sources

Grounded assistants follow a pattern called retrieval-augmented generation (RAG): instead of answering from memory, the model is handed the relevant parts of your documents and writes from them. Google's own description of its tool matches that pattern. When a notebook contains many sources, Gemini Notebook retrieves the most relevant information for your question, then builds its response from it.

In practice, four things happen:

1. Your sources are stored in the notebook. Google defines a source as a copy of the document you upload, or an auto-synced version for Google Drive files.
2. You choose what chat reads. It uses every source, or only the ones you select in the Sources panel. Notes are used only when you select them, and the conversation history is used too.
3. The model answers from what it retrieved. Since October 2025, chat has used Gemini's full 1 million token context window on all plans, which Google said improved its handling of large document collections.
4. Each point carries a citation. Hover over a citation to see the full quoted text; select it to open the quote in context.

Two consequences follow. First, the question matters: Google advises naming the source you mean, for example asking for the key findings in one specific document rather than a general summary, because a specific question helps it find the most relevant passages. Second, each notebook is independent. Gemini Notebook cannot draw on several notebooks at once, so a policy in one notebook does not exist for a question asked in another.

The boundary also depends on where you open the notebook. Inside Gemini Notebook, responses are grounded only in your notebook's sources. The same notebook in the Gemini app or in AI Mode in Google Search is grounded in your sources but may also use web search and other tools. When an answer must come only from your documents, ask it in Gemini Notebook.

Since June 2026, starting with the top plans, chat can also act for you: search the web for new sources, write and run code on a secure cloud computer, and produce files such as spreadsheets, slide decks and charts. Google says you stay in control of which sources join the notebook, labels these functions experimental, and asks you to supervise them and double-check the results.

What you can add as a source, and the limits

As of September 2026, Gemini Notebook accepts these source types:

Each source can hold up to 500,000 words, or 200 MB for an uploaded file, and there is no page limit. A free account can keep up to 100 notebooks with 50 sources each; paid plans raise both (see the plans section below). A YouTube source is removed within 30 days if the video is deleted or made private.

Sources imported from Google Drive sync every few minutes, so the next answer reads the current version of a policy or a handbook. If you lose access to the file, or it is deleted, the source becomes inaccessible: it still counts toward your limit, but chat and Studio stop referring to it. Gemini Notebook cannot edit or delete your original files.

If you do not have sources yet, type a question into the search box in the Sources panel. Fast Research returns candidate web pages or Drive files to review and import. Deep Research browses up to hundreds of websites on your behalf, writes a multi-page report and lets you import the report with its sources. Both give you material to check, not conclusions: web sources vary in quality, so look at who published each one before you keep it.

How to use NotebookLM: your first notebook

1. Sign in with the right account. Go to notebook.google.com, or install the Android or iOS app. It is free for anyone above the age of consent in a supported country. For work material, use your work account, because the data terms follow the account (see the privacy section).
2. Create one notebook per project. Select Create new notebook. Keep a project's material together, since a notebook cannot see another notebook's sources.
3. Add sources. Upload files, pick documents from Google Drive, paste URLs (separated by spaces or new lines) or paste text. With five or more sources, the notebook can label and group them by topic.
4. Read the overview, then ask specific questions. The Chat panel opens with a summary of all your sources. Ask about one topic at a time, name the document when you mean one, and untick sources that are not relevant.
5. Open the citations. Hover to read the quote, select it to see the passage in context. Do this for every figure, date or name you will reuse.
6. Set the chat style if needed. Under Configure Chat, choose Default, Learning Guide or a Custom role and goal, plus a response length.
7. Save and create. Save useful answers with Save to note, then use the Studio panel for a briefing document, study guide, FAQ, Audio Overview or data table. Reports export to Google Docs and tables to Google Sheets.
8. Share deliberately. Add people as Viewer (read-only access to sources and notes) or Editor (can add or remove sources and share further). A personal account can share a notebook with up to 50 people, and each person's chat history stays private to them.

What NotebookLM can make from your sources

The Studio panel turns a notebook's sources into other formats. As of September 2026 it offers:

• Reports: a briefing document, study guide, FAQ or a format you describe, exportable to Google Docs.
• Audio Overviews: AI hosts discussing your sources in four formats: Deep Dive (two hosts, the default), Brief (one speaker, under two minutes), Critique and Debate, in shorter, default or longer lengths.
• Video Overviews: Explainer, Short (about 60 seconds) or Cinematic (English only, 18 and over); a video can take more than 30 minutes to generate.
• Data Tables: rows and columns you describe, exported to Google Sheets with the citations on a second tab. Google's own example is meeting transcripts turned into action items by owner and priority.
• Mind maps, flashcards and quizzes, slide decks and infographics: slide decks download as PDF and infographics as PNG.

Every Studio output is generated, not extracted. Google describes an Audio Overview as a reflection of your sources rather than the hosts' opinions, and warns that the voices and content may contain inaccuracies. Outputs made with Veo, Omni or Nano Banana carry an invisible SynthID watermark, and generated images and videos get a visible one, which some paid plans can switch off. Use the formats to get oriented or to brief others, then check the source for anything that matters.

NotebookLM for business: where it fits

The tool is strongest where a team keeps asking questions that a known set of documents already answers. Google's own Workspace announcement describes sales teams loading product roadmaps, competitor analysis and customer interviews to prepare account plans, and marketers summarizing customer research into campaign briefs.

For sharing inside a company, three details matter. Public sharing is only enabled for consumer accounts; it is turned off for Workspace Enterprise and Education accounts. A "chat view" link hides sources for a focused experience, but Google notes it does not remove a viewer's underlying access, so it is not a way to keep documents from someone. And once a notebook is shared with at least four other people and has had chat activity in the past week, you can see its last seven days of analytics: users and queries per day.

How Google handles your data: personal and work accounts

Google's wording differs by account type, and the difference is in the details.

Personal Google accounts. Google says the content in a notebook will not be used to directly train its foundational AI models unless you choose to give feedback. A thumbs up or thumbs down is that choice: Google collects the associated content (your prompts, sources, uploads and outputs, including any Gemini chats in the notebook), has specially trained teams review it after disconnecting it from your Google Account, and keeps it for up to three years. Data the notebook shares with other Google services, such as the Gemini app, follows those services' privacy notices, including their use for product improvement. Chats with a notebook in AI Mode in Search may be used for ads, personalization and improving Google's generative AI models, while the notebook's sources and instructions are not.

Workspace accounts where Gemini Notebook is a core service. That covers editions such as Business Starter, Standard and Plus, Enterprise Standard and Plus, Frontline and the Education editions. Google says uploaded files, chats and model outputs are not reviewed by human reviewers or used to improve generative AI models, even when you give feedback, and the Google Cloud Terms of Service (or the Workspace for Education terms) apply. Where Gemini Notebook is only an additional service, as with Workspace Individual or Essentials Starter, the ordinary Google Terms of Service and the personal-account wording apply.

Google Cloud. Organizations can also license it through Google Cloud, as Gemini Notebook for Enterprise or within Gemini Enterprise plans. Google says uploaded files remain in your Google Cloud project, data regionalization is honoured, and VPC Service Controls and IAM controls are available.

Plans and usage limits as of September 2026

Google marks every figure here as subject to change.

On personal accounts, limits since September 2, 2026 are measured in compute rather than counted per feature. They take into account prompt complexity, chat length, the number of sources and the features you use, refresh every five hours until you reach a weekly limit, and are two times higher on Google AI Plus and four times higher on Pro. When you run out, Studio outputs can be set to Generate later. Workspace editions still list daily counts: on Business Starter, for example, 50 chats and 3 Audio Overviews a day; on Business Standard and Plus, 500 chats and 20 Audio Overviews.

On a work account, the Workspace administrator turns the service on or off. If a work account cannot open Gemini Notebook, Google's help sends you to that administrator.

Where NotebookLM gets things wrong, and how to check it

Source grounding narrows what the model draws on; it does not make the model read correctly. Google said as much in its first announcement: grounding seemed to reduce the risk of hallucinations, but it was still important to fact-check responses against the original sources. Its May 2026 I/O notebook carried the same reminder that, like all AI, NotebookLM can generate inaccuracies.

The failures to expect:

• A misread passage. A citation proves where the model looked, not that it drew the right conclusion from the passage.
• A missed passage. In a large notebook the model works from the passages it judges most relevant, and a vague question can miss the one that matters.
• Content that never arrived. Footnotes and comments in Google files, images on web pages and everything visual in a YouTube video (only the transcript is imported) are not in the notebook.
• Whole-document citations. When a source is very short, the notebook cites the whole document instead of individual passages.
• Generated media. Audio and Video Overviews can contain inaccuracies and glitches, and are summaries, not records.
• Agent actions. Google says the code, web and file features can make mistakes and do unexpected things.

Google also says not to rely on Gemini Notebook for medical, legal, financial or other professional advice. A workable routine before an answer goes into anything you send:

1. Open the citation behind every number, date, name and obligation you will reuse.
2. Read the cited passage in context, including the sentences around it.
3. Ask which source supports any claim that has no citation; if none does, treat the claim as unsupported.
4. For comparisons across documents, ask for a data table and check the citations cell by cell.
5. Keep the notebook current: remove superseded versions so the old policy and the new one never both answer.

When a notebook is not enough

Gemini Notebook is a tool for people working with documents in Google's own apps. If the questions come from customers on your website, if answers must include live facts such as prices or availability, or if the assistant has to run inside your own systems under your own access rules, that is a different build. Our AI and automation service builds assistants grounded in your content that answer from your own pages, documents and live data, cite their sources and say when they do not know. Sometimes the answer is a feature you already pay for; if it is, we will say so and not build it. For the wider question of how AI gets built into software, see our guide to intelligent apps.