NotebookLM Guide

Source-grounded research: Google’s answer to the “hallucination reflex”

NotebookLM is Google’s source-grounded research and synthesis assistant. You load your own sources (PDFs, Google Docs, websites, YouTube videos, audio, EPUBs and so on) into a Notebook, and NotebookLM answers questions, produces briefings, Audio Overviews in podcast format, Video Overviews, Mind Maps, quizzes, Slide Decks and Infographics. All of it based exclusively on the sources you supply, with inline citations.

Unlike generic chat models, every answer is grounded in your Sources (a RAG architecture over a long context window). Since December 2025 the backend has run on Gemini 3, and since 8 June 2026 on Gemini 3.5 with the Antigravity framework (initially for Ultra and Workspace AI Ultra/Expanded Access accounts, from 16 July 2026 also for Google AI Pro on the web; that rollout has been complete since 5 August 2026). Since April 2026, Notebooks in Gemini are synchronised bidirectionally (since 17 April 2026 also for free users on gemini.google.com). That makes NotebookLM more than a Q&A tool: it is a central knowledge workbench that produces Studio outputs (audio, video, Mind Map, reports, Slide Decks, quizzes) as reusable artefacts.

Backend & multimodality

• December 2025: Gemini 3 replaces Gemini 2.5 Flash. Multimodal understanding, stronger reasoning, 1 million token context window even on the free tier.
• 8 June 2026: Gemini 3.5 with the Antigravity framework becomes the default model: higher accuracy, better grounding fidelity and transparent reasoning steps in chat. This turns NotebookLM from a pure RAG system into an agentic research workbench: every Notebook gets a secure cloud computer with more than 100 curated software skills that write and execute code in the background (data transformation, statistical analysis), with no manual configuration. Rolled out initially only to Google AI Ultra and Workspace Business customers with AI Ultra Access or AI Expanded Access; from 16 July 2026 the upgrade (incl. cloud computer and code execution) also rolled out to Google AI Pro on the web. That rollout has been complete since 5 August 2026, so code execution as well as PDF, spreadsheet and chart outputs are now open to all AI Pro accounts. The free tier remains excluded.
• Cinematic Video Overviews (since March 2026): additionally powered by Nano Banana Pro (images) and Veo 3 (video).

Tier overview (as of May 2026)

Feature timeline at a glance

When is NotebookLM worth it, and when not?

What is a Notebook?

A Notebook is a self-contained workspace for one topic or project. It consists of three main areas that sit side by side in the browser:

1. Sources panel (left)

All sources you have uploaded or imported via Discover Sources / Deep Research. A maximum of 50 (Free) · 100 (Plus) · 300 (Pro) · 500/600 (Ultra 20 TB/30 TB) per Notebook. Each source can be enabled or disabled individually per question, which is the basis for selective sourcing (see Section 5).

2. Chat panel (centre)

Your dialogue with the model. Answers contain inline citations with a jump link to the original source. Since 2026 with goal setting: you tell the Notebook your goal (essay, briefing, comparison, pitch) and the answers align with it consistently.

3. Studio panel (right)

All generated artefacts: Audio Overview, Video Overview, Mind Map, briefing, Quiz, Slide Deck, Infographic, Data Table, timeline. Since July 2025, multiple outputs of the same type per Notebook are possible, so you can keep old and new versions of a briefing side by side.

How does NotebookLM ground its answers?

At its core, NotebookLM is a RAG system with a long context window. When you upload sources, they are embedded and stored. During chat, Gemini 3 pulls the relevant source passages into the 1 million token context window (on all tiers since December 2025), cites from them and inserts evidence. The answers are thus verifiably grounded in your Sources.

Three terms you should keep cleanly apart

• Source: a single uploaded source (PDF, Doc, YouTube…) in the Notebook.
• Note: an answer saved from the chat (“Save to Note”), which can itself become a Source again.
• Studio output: a generated artefact in the Studio panel (audio, video, Mind Map …).

There are six ways to start a Notebook, add Sources or pull content from a Notebook into a flow. Not all of them are equally available in every tier.

1. Web app (notebook.google.com)

The new primary address since late July 2026. The old link notebooklm.google.com redirects there automatically, and saved bookmarks keep working.

1. Sign in with your Google account (personal account or Workspace).
2. “Create new notebook” takes you straight to the upload modal.
3. Add sources: drag and drop PDFs, pick files from Drive, paste a URL, add a YouTube link, use Discover Sources or Deep Research.
4. As soon as at least one source is loaded, the Chat and Studio panels unlock.

2. Mobile app (iOS / Android, since May 2025)

• System share sheet: In your browser, YouTube or Drive tap “Share” → NotebookLM → the source lands directly in the Notebook you choose.
• Offline download and background playback for Audio Overviews, ideal for your commute.
• “Join” feature (beta): join a running Audio Overview live and talk to the hosts.
• Real-time voice chat (since 15 September 2026): Instead of typing, you talk to the Notebook in almost 100 languages, interrupting mid-sentence and getting step-by-step explanations along your Sources. Launching on Google AI Ultra (accounts aged 18+ only); since 29 September 2026 fully available for Google AI Pro as well, according to Google. Android and iOS.
• Recorder (rolling out from the week of 22 September 2026): Record directly in the app: a lecture, a thought from a meeting, a voice note. The recording lands alongside your other Sources in the Notebook and can be cited and edited from then on. Google explicitly includes accounts under 18 with English as the output language.
• Requirement: iOS 17+ or Android 10+.

3. Workspace Drive integration

• In Google Docs: “Help me write” / Gemini sidebar → “Open in NotebookLM” creates a Notebook with the Doc as its first source.
• In Drive: right-click a file → “Open with NotebookLM” (for supported formats).

4. Workspace Studio “Ask NotebookLM” since 12 May 2026

Workspace customers now get a new step called “Ask NotebookLM” in Workspace Studio (Google's automation platform for Workspace flows). It lets you plug an existing Notebook into automations as a grounded knowledge source, e.g. to answer tickets, emails or briefings based on curated Sources. Available for Business Starter/Standard/Plus, Enterprise Standard/Plus, Education Fundamentals/Standard/Plus as well as Google AI Pro for Education, Teaching & Learning and AI Expanded Access. Admins control it via Gemini for Workspace step permissions. Since 1 June 2026, licences with AI Expanded Access have higher usage limits for Workspace Studio.

4b. Workspace Studio “Add a source to Gemini Notebook” since 6 August 2026

The counterpart to “Ask NotebookLM”: instead of reading from a Notebook, this step writes new Sources into it. Supported are text, links to Drive files and YouTube and generic web URLs. That lets you run a Notebook as a living knowledge base that keeps itself current via a flow, for instance a weekly market radar that automatically collects new industry articles, or an onboarding Notebook that picks up every newly shared process document from Drive.

Available for Business Starter/Standard/Plus, Enterprise Standard/Plus and Education Fundamentals/Standard/Plus. The step is enabled by default as long as the organisation allows Gemini for Workspace steps. Rolling out from 6 August 2026 with up to 15 business days of lead time; personal Google accounts are excluded.

5. Notebooks in Gemini since April 2026

Create a new “Notebook” in the Gemini app and it appears bidirectionally in NotebookLM as well. You can add Sources in Gemini and generate Studio outputs in NotebookLM without switching interfaces. The Gemini web sidebar gets a new “Notebooks” section above Gems and Chats.

6. Notebooks in the AI Mode of Google Search since 19 August 2026

The AI Mode integration announced in July has arrived: your existing Notebooks from Gemini Notebook and the Gemini app appear automatically in Google Search, without you having to recreate them there. Sources and Instructions are synchronised per Notebook across products. In AI Mode you pick a Notebook in the sidebar and ask your question against exactly its sources instead of the open web. You can additionally add earlier AI Mode threads as a Source to a Notebook and have finished documents generated from uploaded files or threads.

The rollout has been running since 19 August 2026 over a few days in more than 180 countries in English; further languages follow in the weeks after.

7. Google Docs: a Notebook as context via @ since 23 September 2026

In Google Docs, type @ in the Gemini side panel or the bottom bar and pick an existing Notebook. Gemini then writes grounded in that Notebook’s Sources and adds inline citations, so you can check facts without switching tabs. The Notebook can be combined with other context sources such as emails, chats and Drive files. Available for Business and Enterprise Standard/Plus, Education Plus, Google AI Pro/Ultra, Google AI Pro for Education and AI Expanded Access, on Rapid and Scheduled Release.

Notebook limits per tier

NotebookLM has two categories of building blocks: input blocks (Sources), i.e. what you pour in, and output blocks (Studio), i.e. what comes out.

A) Source types

Expert Intelligence: purchased books as a Source since 27 August 2026

With Expert Intelligence, Google opens up a licensed publisher corpus for Gemini Notebook for the first time: e-books you have bought in Google Play Books can be loaded directly into a Notebook as a Source. In Play Books, every eligible title carries a “Tools” badge; one click pushes the book into the Notebook, and from then on you chat grounded on the book's content.

• Scope at launch: more than 100,000 titles from Bloomsbury, De Gruyter Brill, Johns Hopkins University Press, Macmillan, O’Reilly Media and Penguin Random House.
• Purchased titles only: there is no flat rate and no catalogue access. What you do not own, you cannot load.
• Announced: expansion to the Gemini app and the AI Mode of Search, as well as to third-party subscriptions, business research reports and textbooks.

B) Studio output types

Audio Overview (podcast)

• What: Two AI hosts discuss your Sources in a 10 to 25 minute podcast.
• Customize modes: Deep Dive (default), The Brief (quick overview), The Critique (critical questioning), The Debate (pro/con). Plus topic focus, target audience and expertise level freely via prompt.
• Lengths: Shorter / Default / Longer. “Longer” is currently only available in English; if you want German, you have to stay with Default or Shorter.
• Languages: 80+ incl. German (expanded to 50+ in April 2025, to 80+ in July 2025).
• Daily limits: Free 3, Plus 20, Ultra 200.
• Interactive Mode: “Join”, where you talk to the hosts live.

Video Overview

• Explainer (default): Slide-style narrated video (3 to 6 minutes) with diagrams, images and quotes from your sources. Supports 50+ languages, incl. German.
• Short Video Overview (since early July 2026): A roughly 60-second video in portrait format (9:16, “TikTok style”) that nails the one core idea of a source. Powered by Nano Banana 2 Lite. Rolling out on web and mobile, initially for AI Pro and Ultra, with free users to follow. Available in more than 80 languages since 15 September 2026, and German is among them. Google combines the narrated summary with explanatory animations (intended for formulas, diagrams, complex concepts); the rollout runs over several weeks. That leaves only Cinematic as English-only.
• Brief: The short counterpart to Explainer: a compact video overview that only makes the core ideas of your sources tangible instead of working through them in a structured way. Available in all supported languages, so in German too. Not to be confused with the Short Video Overview: Brief is a format in landscape orientation, Short is a separate portrait-format output.
• Cinematic Video Overview (since March 2026): Animated, story-driven videos with Nano Banana Pro + Veo 3. English only.
• Visual Styles: Explainer and Brief are illustrated with Nano Banana and can be set to six styles: Watercolor, Papercraft, Anime, Whiteboard, Retro Print and Heritage. Without a selection, NotebookLM picks a suitable style itself. For client assets a deliberate choice pays off: Whiteboard looks more serious in a B2B briefing than Anime.
• Customize: format (Explainer / Brief), Visual Style, topics, learning objectives, target audience.
• Daily limits: Free limited, Ultra 200.

Mind Map

• What: Branching diagram of the main topics and subtopics from your Sources. Every node is clickable → opens a chat question about that node.
• Export: PNG download or via sharing the Notebook.

Briefing Doc

Condensed summary with main points, findings and action items. 1 to 3 pages. The default output for “What is in there?”

Study Guide

Outline, quiz with answer key, essay questions, glossary. The ideal output for teaching and onboarding.

FAQ

Q&A list from the Sources, ideal for a knowledge base or support page.

Timeline

Chronological sequence of events / milestones. Suited to histories, project phases, contract timelines.

Quiz / Practice Tests

Multiple-choice + open-ended questions from the Sources. Very useful in training contexts (e.g. AZAV courses). Since March 2026 with progress preservation: progress is saved per session, “Got it”/“Missed it” markers, shuffle function, results screen with missed items. Via the pencil icon you can set the difficulty (easy / medium / hard), the quantity (fewer / standard / more) and a free-form prompt before generating.

New question types (announced 15 September 2026, rolling out over the following weeks): in addition to multiple choice come Short Answer, Multiple Select and Fill-in-the-Blank. Also new is the follow-up chat: after completing a quiz or flashcard round, you ask the Notebook in chat about your performance and get back where the gaps are. For training this means: the test evaluation step you used to build yourself is now inside the tool.

Flashcards expanded since March 2026

Flashcards from the Sources with mark-as-learned logic, shuffle and deletable individual cards. Progress is kept between sessions.

Learning Overview since 15 September 2026

The first interactive output type: a Learning Overview lives under Reports and bundles what previously sat side by side as loose individual outputs: Summary, Infographic, Quiz and Flashcards in one document that you click through instead of scrolling. Designed for focused learning on a single topic.

• Availability: according to Google for all users, rolling out over the weeks after 15 September 2026. Google has not mentioned any tier restriction.
• Why this matters for enablement: A Learning Overview comes close to a finished learning unit: explanation, visual and check in one artefact. For onboarding tracks and training material it replaces the previous three-step routine of Briefing Doc, infographic and separate quiz.
• Limit: It remains a generated output over your Sources; the didactic sequence, the weighting and the question of what should be tested at all are still yours to decide. Before using it in a publicly funded training programme (AZAV), every generated quiz needs to be proofread.

Slide Deck since March 2026

• What: Generated presentation deck from your Sources (with a visible watermark; on Google AI Pro and Ultra it can be switched off since 14 August 2026).
• Customize: prompt-based slide revisions (“Make slide 4 a comparison table”) on desktop and mobile.
• Export: since March 2026 PPTX export (PowerPoint) in addition to Google Slides export.

Infographic since March 2026

Image infographic from the most important data points / concepts. With a visible watermark; on Google AI Pro and Ultra it can be switched off since 14 August 2026. Since March 2026 there are 10 manually selectable styles: Sketch Note, Kawaii, Professional, Scientific, Anime, Clay, Editorial, Instructional, Bento Grid, Bricks.

Data Tables since December 2025

Extracts structured tables from the Sources: comparable, editable. Very strong for clause comparisons, KPI tables, feature matrices.

Generated Notes

Your own notes that you save in chat via “Save to Note”; they then become a Source in the Notebook themselves. This way your Notebook grows organically into a knowledge pyramid.

Agentic Studio & free download formats since June 2026

With Gemini 3.5 + Antigravity, the Studio no longer produces only the fixed output types but freely instructable, fully formatted and editable files. You describe what you need (“a budget sheet with quarterly columns”, “an 8-page briefing as a PDF”); NotebookLM pulls the context from your Sources, runs code in the background where needed, and returns the result as a download. After generation, further changes are possible via chat.

• Documents: PDF, Word (.docx), Markdown, plain text
• Presentations: PowerPoint (.pptx)
• Data & tables: Excel (.xlsx), CSV, JSON
• Visuals & charts: data visualisations as PNG / SVG, images as PNG / JPG / GIF (via Nano Banana)

Google has announced further formats. Availability: initially only Google AI Ultra + Workspace Business with AI Ultra/Expanded Access (web); free tier excluded for now, and the mobile app was not included at launch.

Chat-driven source building since June 2026

You no longer have to upload all sources up front: start a chat about your project and NotebookLM suggests suitable sources via Google Search and its own research skills, including primary sources in other languages and material by related authors, and adds them straight into the Notebook on request. An extension of the existing Discover Sources logic.

Overview table: output type vs. use case

Source pinning / Selective Sourcing

Every Source has a checkbox in the Sources panel. Per question you can activate only a selection, e.g. “Answer this only from Sources 3 and 7”. Select all / Deselect all lets you switch quickly between “everything” and “focused”.

Custom Instructions since February 2026

Up to 10,000 characters per Notebook, freely configurable: tone, language, goal, persona of the model. Sent automatically with every question, saving you prompt overhead.

Always answer in English and address me informally.
Cite as [Source X, page Y].
Flag uncertainties explicitly ("cannot be clearly derived from the sources").
Keep a factual, advisory tone -
no marketing language, no filler words.

You are my sales researcher. I am preparing a pitch for [client name].
For every question, deliver:
 1. A direct answer (2-3 sentences)
 2. Source evidence with [Source X]
 3. Three possible pain points / conversation hooks
Language: English. Tone: factual, advisory.

Customize options per Studio output

“Save to Note” / “Convert to Note”

Any chat answer can be saved as a Note with one click. Notes then become a Source in their own right, so you can build on earlier insights without asking the question again. Over time the Notebook turns into a knowledge pyramid: raw Sources at the bottom, Notes as a condensed intermediate layer in the middle, Studio outputs as final artefacts at the top.

How context is built for questions

• Default: If you deactivate nothing, Gemini 3 searches all activated Sources in the 1 million token context window.
• Selective: If only 3 of 50 Sources are activated, the model sees only those 3, perfect for targeted cross-checks.
• Citations: Answers always contain inline markers [1] [2] [3] that jump to the original passage on click.
• Goal-Setting (2026): You can give the Notebook a Goal (“I'm preparing a pitch for ACME”), after which answers are consistently optimised for it.

Permission levels

Sharing models

1. Person-specific (by email): As in Google Docs: add an email address, choose a role.
2. Link sharing: Generate a link for anyone with a Google account.
3. Public Notebooks: Make a Notebook public → anyone with a Google account can read, chat and consume the pre-generated outputs, but not add Sources. Since launch (May 2025) more than 140,000 public Notebooks have been created.
4. Featured Notebooks: Curated Notebooks from Google partners (The Economist, The Atlantic, research institutes, the works of Shakespeare etc.). Visible in the “Featured notebooks” tab.

Workspace integration

• Personal account ↔ Workspace account: Cross-domain sharing is possible with restrictions, but not all features (e.g. editing) work. Recommendation: keep everyone involved in the same Workspace tenant.
• Within Workspace: Sharing defaults follow the domain policies. Since 10 September 2026, admins no longer control external sharing via a single on/off switch but in four levels; see below.

External sharing in the Admin console since 10 September 2026

Google has replaced the previous simple on/off switch for external sharing with four graduated settings. They can be set at domain, OU and group level, so the marketing department may share publicly while HR and Finance cannot share outside the domain at all.

Rolling out from 10 September 2026 to all Workspace editions, Rapid and Scheduled Release simultaneously, with up to 15 days until it becomes visible.

Important for the order of your governance work: this setting governs with whom sharing is allowed, not what travels along. The “Leak via Synthesis” above and the “Allow copies” switch, which is on by default, are unaffected and remain with the person who owns the Notebook.

Permissions at Source level

A Source remains subject to its own permissions. If you add a Google Doc as a Source that others are not allowed to see, recipients of the Notebook get no access to the original Doc, but they do see the content and citations the Notebook generates from it.

Copying a Notebook since 17 August 2026

A Notebook can now be pulled into your own library as a complete copy: the clean way to derive your own working version from a shared team Notebook without touching the original.

• Drive Sources only with your own access: A Drive-based Source only lands in the copy if the person copying may view and copy the original file. Otherwise it silently drops out of the copy, and the Notebook is incomplete afterwards without anyone being told.
• No sync connection: The copy is a snapshot. If the person who owns the original changes something later, none of it flows into the copy.
• Control lies with the owner: The sharing settings contain the switch “Allow copies”, and it is on by default.
• Availability: all Workspace customers and personal accounts with Gemini Notebook access, Rapid and Scheduled Release immediately. There is no dedicated admin setting for it; admins can only switch Gemini Notebook on or off as a whole.

Real-time collaboration

Several editors can work simultaneously: Sources are shared live and Studio outputs are visible to everyone. Chat history, however, is individual per user; everyone sees their own question history.

Enterprise: its own sharing rules

In Gemini Notebook Enterprise (GCP), tighter rules apply than in the Workspace or personal variant. You can share only with people who are in the same project, hold the Cloud NotebookLM User role, belong to the same Workforce Pool and have a Gemini Notebook Enterprise or Gemini Enterprise licence. It stays with the two roles Viewer and Editor, although editors here may neither delete nor share nor revoke access.

• Autocomplete when sharing: Email addresses and group names are suggested in the sharing dialog. With Cloud Identity this works out of the box; with Workforce Identity Federation only if SCIM is configured. New since 20 August 2026: Autocomplete is now also available (GA) with Okta as the external identity provider, provided Okta is connected via Workforce Identity Federation and SCIM is configured with extended attributes. Without SCIM you still have to type addresses in full.
• Two hard limits: A Notebook that uses Gemini Enterprise search results as a Source cannot be shared at all. And once a Notebook is shared, no new Gemini Enterprise Sources can be added to it. If you need both, plan the order: curate Sources first, then share.

Plus / Enterprise: Usage Analytics

NotebookLM Plus and Enterprise offer source read analytics: you can see which Sources were read how often. Very useful for measuring the success of team Notebooks (onboarding, knowledge base).

Data usage: the most important question

Free / personal account

Google puts it this way: “The content in NotebookLM will not be used to directly train our foundational AI models, unless you choose to provide feedback.”

By default there is no training on your data. But: if you click thumbs up/thumbs down or give feedback, the content can be reviewed by humans (disconnected from your Google account, retained for up to 3 years).

Workspace / Enterprise

Google puts it this way: “Your uploads, queries and the model's responses in NotebookLM will not be reviewed by human reviewers even when you provide thumbs up or down feedback, and will not be used to train AI models.”

Strongest protection. No training, no human review, not even when you give feedback. The standard Workspace terms apply.

Free vs. Plus: any difference in data handling?

NotebookLM Plus, where it is included in a personal Google AI Plus/Pro/Ultra plan, runs under the Google personal terms: the same feedback rules as Free. Only when NotebookLM Plus is obtained through Google Workspace do the enterprise-grade protections apply.

Enterprise via Google Cloud / Gemini for Workspace

• NotebookLM Enterprise runs in the customer's own GCP project. Data never leaves your own GCP tenant, is VPC-SC compatible and supports CMEK (Customer-Managed Encryption Keys via Cloud KMS).
• VPC-SC excludes website URLs (since 9 September 2026): As soon as VPC Service Controls are active for a project, website URLs can no longer be added as a Source. The reason lies in the mechanism itself: direct website ingestion runs a live crawl and thereby generates outbound traffic beyond Google's networks, which is exactly what the perimeter is meant to prevent. Other Source types remain usable, explicitly including Google Docs and YouTube URLs. In practice for DACH: the strictest security perimeter costs you the most convenient research path. If you still need web content, save it as a PDF or Google Doc in Drive and add it as a Source from there. Plan that copying step as a deliberate, documented action, because the provenance no longer travels with it automatically.
• Data residency: multi-region eu (EU) and us (US) are generally available; in-country regions for Canada and India are in restricted GA (allowlist via Google Sales). As of May 2026 there is no Germany/Frankfurt in-country region: EU data residency runs through the eu multi-region. For DPA purposes and EU-only guarantees, always obtain binding confirmation in the DPA.
• DRZ/MLP in the EU (since 3 June 2026): Gemini Enterprise meets the data residency (DRZ) and Machine Learning Processing (MLP) requirements in the EU for NotebookLM Enterprise, but only for adding Sources and chatting with those Sources. Discover Sources and all Studio outputs are not MLP-compliant (Audio Overview, Video Overview, Slide Deck, Infographic, Mind Map, Reports). The practical consequence for DACH: if you have to guarantee EU processing contractually, you can use Sources and chat, but you must keep the entire Studio locked for regulated content. This boundary cuts right across the product and is not a tier question: even Enterprise does not automatically protect everything here.
• IAM: access via the Cloud NotebookLM User role (employees) and the Cloud NotebookLM Admin role (identity/user management), controllable granularly per project. Authentication via Cloud Identity or third-party providers (Microsoft Entra ID, Okta).
• Audit logs: Since February 2026, usage audit logs (user queries + responses) have been available in Cloud Logging.
• Compliance: Since 31 March 2026, NotebookLM Enterprise has been BSI C5:2020 certified, which is relevant for DACH customers in regulated industries (the BSI's Cloud Computing Compliance Catalogue).
• Slide Decks & Infographics have also been generally available in NotebookLM Enterprise since 19 May 2026.
• Sensitive Data Protection for Sources (GA since 31 August 2026): You can apply Content Policies from Cloud DLP to Gemini Notebook Enterprise notebooks. The policy checks a Source before it enters the Notebook and blocks the upload if the content violates the policy; the Source then appears in the list with a violation notice rather than being masked or flagged afterwards. Prerequisites: the roles/dlp.user role for the Gemini Enterprise service account and a policy in the same region as the notebooks (global, us or eu). Know the limits: audio files, Google Drive documents and Slides are not checked, and very large files are let through by default rather than scanned. So the safeguard is real but not watertight: it is no substitute for training people on which data belongs in a Notebook in the first place.
• A note on naming: Since the rebrand on 16 July 2026, the product is called Gemini Notebook Enterprise in the Cloud documentation. Feature scope and API endpoints remain unchanged; subscription pages and IAM role names (Cloud NotebookLM User / Admin) still partly show the old name.

Workspace Core Service since February 2025

NotebookLM and NotebookLM Plus are officially Core Services in Google Workspace, i.e. the same data protection and support standards as Gmail, Docs and Drive.

Editions included:

• NotebookLM (Standard): Business Starter, Frontline Starter/Standard, Essentials, Nonprofits
• NotebookLM Plus: Business Standard/Plus, Enterprise Standard/Plus, buyers of the Gemini Business/Enterprise add-on

Admins can enable or disable NotebookLM per user via the Admin console.

Audit logs in the Workspace Admin console since 3 September 2026

Until then, audit logs existed only on the Cloud side (Gemini Notebook Enterprise via Cloud Logging). Since 3 September 2026, Google has been rolling out comprehensive audit logs for Gemini Notebook to the Workspace Admin console (gradually over up to 15 days, Rapid and Scheduled Release simultaneously). For the first time, a Workspace admin can see what happens in a Notebook without GCP.

• Where: in the Security Investigation Tool and the Audit and Investigation Tool.
• What: a broad set of user actions across several categories, including Notebook visibility, user identity, IP address and resource context.
• For whom: all Workspace customers with access to these two tools.
• Default: the logs are there out of the box. The export to BigQuery, by contrast, is switched off until an admin actively enables it.

Core Service does not mean data region

What CANNOT be uploaded?

• Files > 200 MB or > 500,000 words per Source.
• Paywalled content (websites behind a subscription barrier).
• Private YouTube videos or videos < 72 h old.
• Embedded videos / nested web pages.
• Heavily scanned / image-based PDFs without an OCR layer.
• Sensitive PHI / health data: not explicitly prohibited, but excluded by many compliance frameworks, especially in the Free version with its feedback risk.

API availability

This section is aimed at anyone who wants to use NotebookLM with customer data, contracts, internal strategies or other sensitive content. Anyone who “just quickly” sets up a personal Notebook is risking more than they would like.

The decisive distinction: personal plan vs. Workspace

Checklist for DACH SMEs

1. Create Notebooks only in the Workspace account. Clear policy: company Sources may only go into the company NotebookLM, never into an employee's personal account.
2. Data processing agreement (DPA) with Google or the Workspace reseller: have it reviewed. For Enterprise via GCP, additionally set the EU region and have it confirmed in the DPA.
3. Configure the Admin console: enable NotebookLM only for employees who have completed compliance training. For external sharing, since 10 September 2026 choose the appropriate one of the four levels (Off / Trusted Domains / On / On with public notebook sharing), sensibly per OU, and allow public notebooks only where they are genuinely needed.
4. Settle the recorder before rollout. Since September 2026, the mobile app records lectures and conversations and stores them as a Source in the Notebook. In Germany, recording conversations requires the consent of everyone involved; on top of that, define retention and purpose and involve the works council (Betriebsrat) before the feature lands on company devices.
5. Check the data region, and plan for the gap. An EU data region in Workspace does not cover Gemini Notebook: Notebooks, Sources and chat histories are stored globally. If you have to guarantee EU storage, you need Enterprise via GCP, or a documented decision about which content must therefore never go into a Notebook in the first place.
6. Introduce audit logs deliberately. Since September 2026, the Workspace Admin console has logged Notebook actions including user identity and IP address. That is a gain for traceability and at the same time employee behavioural data: involve the works council, define purpose limitation and retention, and only switch on the BigQuery export afterwards.
7. Contracts and PHI only in the Workspace/Enterprise tier. Never in the Free tier with its feedback risk.
8. Raise employee awareness: thumbs up/thumbs down in a personal account can lead to human review. Not in Workspace, but the rule has to be known.

Risk: “personal account in the company”

A common anti-pattern: employees create Notebooks in their personal Google account because it is “quicker” or because they have not been given access to the Workspace version. Then they upload confidential company data, and everything falls under personal terms rather than Workspace protection.

Fix: a clear internal policy, awareness training during onboarding, and a regular reminder at the team all-hands.

Risk: “leak via synthesis”

Even if the Source permissions are correct, Studio outputs can let Source content leak through (see Section 6). For confidential sources, generate and share Studio outputs only in closed Notebooks.

Minimum training

Before anyone on the team works productively with NotebookLM, they should:

• know the difference between a personal account and a Workspace account,
• understand what “a Source remains subject to its own permissions” means,
• know what must NOT be uploaded (PHI, paywalled content, confidential PDFs into a personal account),
• have created a test Notebook themselves and generated an Audio Overview, in order to get a feel for what the tool actually does.

Auto-update when Sources change?

• Google Docs / Slides / Sheets: Since 26 May 2026 these Drive Sources sync automatically: if the original changes in Drive, NotebookLM pulls the new version on its own (no more manual re-adding). Available for all Workspace customers and personal accounts with NotebookLM access, without a separate setting. Permissions are enforced in real time: if you lose access to a Drive file, it can no longer be used as a Source; if the original is deleted, it disappears from the Notebook too. But: Studio outputs already generated (audio, briefing) are not regenerated automatically.
• PDFs / uploaded files / web URLs: No auto-updates. Snapshot at upload time. You have to delete the source and add it again when the original changes.
• YouTube: Snapshot of the transcript at the time it was added.

Version history

• No native version history per Notebook (as of May 2026). Unlike Google Docs there is no File → Version History.
• Chat history is retained per user; since March 2026 it is auto-saved and kept private (“Saved Chat History”).
• Studio outputs: Since July 2025, several outputs of the same type can be kept, so you can hold old and new versions of a briefing side by side.

Export options

In practice: source refresh routine

For Notebooks that need continuously changing data (market data, newsletters, press reviews):

• Use Google Docs as Sources; they are re-read live.
• Create a separate Notebook per cut-off date (e.g. “Competitors Q2 2026”) instead of maintaining one mega Notebook.
• Establish a re-upload routine in the team: once a month / quarter delete old PDFs, add new ones.

Overview of all tiers (limits checked against the Help Center on 17 September 2026)

Per-source limit (the same in all tiers)

• At most 500,000 words OR 200 MB file size per Source.
• For YouTube: the caption file must not exceed 500,000 words, otherwise it is rejected.

In practice: which tier for which use?

1. One question = one Notebook. Avoid building a mega Notebook for everything. Sources on different topics dilute the answers. Better 5 focused Notebooks than one dump.
2. Source hygiene: 5 to 20 high-quality sources beat 50 mediocre ones. The first Sources are often weighted more heavily in the ranking, so add your most important / most current ones first.
3. Use Selective Sourcing actively. For comparative questions, deactivate first, then ask; otherwise the model falls back to the average across all Sources.
4. Set Custom Instructions (10,000 characters available). Tone, language, persona, output format. That saves you prompt overhead on every question.
5. Use Goal-Setting for the Notebook. Tell the Notebook proactively “I am preparing a pitch for ACME” and answers are consistently optimised for it.
6. Customise Audio Overviews, do not accept the default. The default is English and Deep Dive. Set your language plus the right mode (Brief for a quick sync, Critique for internal reviews).
7. Mind Maps for structure, not for truth. A Mind Map shows what is in the Sources, not what is correct. With contradictory Sources the map can be one-sided.
8. Produce Notes as intermediate artefacts. Turn important chat answers into Sources via “Save to Note”; this is how the Notebook grows into a knowledge pyramid.
9. Actively maintain source updates. PDFs and web URLs are snapshots! If your source changes, delete and re-upload. For continuously changing data, prefer Google Doc Sources (live-linked).
10. Confidential workflows only in Workspace NotebookLM. Anyone working with client data, contracts or PHI must obtain NotebookLM via Workspace, not via the personal subscription.

The 7 most common pitfalls

Pitfall 1: “Source mix poisoning”

You load 30 unrelated Sources into one Notebook → answers become generic, Mind Maps messy, Audio Overviews thematically blurred.

Pitfall 2: PDFs with images / tables

Gemini 3 is multimodal, but NotebookLM extracts primarily text from PDFs. Important charts, diagrams or scanned documents are often missed. Fix: Upload important images additionally as separate image Sources, or pre-process the PDF with good OCR.

Pitfall 3: YouTube is transcript only

Only the caption file is imported; visual information from videos (slides, demos) is invisible in the Notebook. Fix: Add the speaker slides separately as a PDF/PPT Source.

Pitfall 4: Trusting the Audio Overview blindly

Audio Overviews sound extremely convincing, but they are synthesised; hallucinations do occur (a rate of roughly 13%). Always cross-check important facts in the citations.

Pitfall 5: The “snapshot instead of live” trap

Uploaded PDFs/EPUBs and web URLs are not auto-updated. Anyone who uses a Notebook for weeks and forgets that a source has since gone stale is building on dusty data. Fix: Establish a source refresh routine, or use Google Docs / Sheets / Slides from Drive as Sources; those have synced automatically since 26 May 2026.

Pitfall 6: Personal account at work

Employees create Notebooks in their personal Google account and upload confidential company data → this falls under Personal Terms, not Workspace protection. Fix: A clear policy: company Sources only in the Workspace account.

Pitfall 7: Cross-Notebook search is impossible

Notebooks are silos. You cannot search several Notebooks at once. Workaround: Consolidate the most important insights as Notes, or try cross-cutting questions via Notebooks in Gemini.

Pitfall 8: Website URL cannot be added in the Enterprise project

In Gemini Notebook Enterprise, adding a website URL fails while Google Docs and YouTube links go through without a problem. Cause: Since 9 September 2026 this is not a bug but a documented limit: in projects with VPC Service Controls, direct website ingestion is blocked because it triggers a live crawl and thereby generates traffic out of the perimeter. Fix: Save the content as a PDF or put it into Drive as a Google Doc and pull it in as a Source from there. Note the source and retrieval date manually in the document, otherwise the provenance is lost. Do not tamper with the perimeter to get the convenience back; that is the protection you chose.

Troubleshooting: top 5 with solutions

1. “Source could not be added”

Possible causes: file > 200 MB, > 500k words, YouTube video without captions, paywalled URL, private Doc without access.

Fix: Split the PDF, enable captions in YouTube Studio, “print” the web page as a PDF and upload it.

2. Answers ignore certain Sources

Cause: Long Sources are weighted differently in the 1 million token context; the first / shorter Sources often dominate.

Fix: Use Selective Sourcing actively and activate only the Sources relevant to the question. Or: split long Sources into topic chunks and upload them individually.

3. Audio Overview is in English although another language is set

Cause: The output language is set at profile level and/or Notebook level; if the two conflict, the Notebook setting wins. The “Longer” variant is English only.

Fix: Settings → Output Language explicitly set to your language, and choose the length option “Default” or “Shorter”.

4. Studio output does not regenerate

Cause: Studio outputs have been multi-version since July 2025; the old output stays, the new one appears next to it.

Fix: Click “+ Create new” in the Studio tile and optionally delete old versions.

5. Sharing link does not work for external users

Cause: A Workspace domain restriction blocks external sharing, or the recipient has no Google account.

Fix: Check the sharing policy with the Workspace admin. Alternatively: make the Notebook Public.