NotebookLM data handling & privacy
What You'll Find Below
A plain-language summary of how Google's AI research notebook handles your uploaded material — covering training opt-outs, retention windows, Workspace versus personal account differences, and how the citation model keeps content within your sources.
Data privacy is the first question many organisations ask before adopting any AI tool, and the research notebook is no exception. The short version: Google states that uploads to the tool are not used to train its foundation models, each notebook is isolated to the account that owns it, and Workspace customers have additional administrative controls. What follows is a more complete picture.
Training opt-out: what Google says
Google's published product terms state that content uploaded to the notebook by personal-account users is not used to train or improve Google's AI models. This is not a setting a user has to find and toggle — it is the default for the product. The same position applies to audio generated by the tool: your audio overviews are not fed back into the voice model's training pipeline.
For Workspace customers, the terms sit inside the Google Cloud Data Processing Addendum rather than the consumer privacy policy. The Addendum likewise excludes customer content from being used to train Google's models. Administrators concerned about compliance can confirm the current addendum language through Google's Workspace admin console or their account representative.
Retention windows and deletion
Sources you upload stay in a notebook for as long as the notebook exists. Removing a source from the panel deletes it from the active index and removes its passages from future answers. That change is near-immediate — subsequent questions will not cite the deleted source. Deleting the entire notebook removes all associated sources, notes, and generated content from the active system.
After account closure or notebook deletion, Google's standard retention and deletion timelines apply. These are set by Google's overall privacy policy and, for Workspace customers, the data-processing agreement in place for that account. The tool does not have separate, longer retention windows that would cause data to persist after a standard deletion.
Personal accounts versus Workspace
The functional experience is largely the same across tiers, but the governance layer differs. Personal accounts are covered by the Google Privacy Policy and the notebook's own terms of service. Workspace accounts have the additional protections of the Cloud Data Processing Addendum, including support for data-residency region selection, admin-level audit logging, and the ability for administrators to disable access for specific organisational units. Educators using Workspace for Education accounts should review Google's specific Workspace for Education terms, which carry additional student-data protections under applicable law.
The FTC business guidance on privacy and security is a useful independent baseline for organisations evaluating whether any AI tool's data commitments are adequate for their sector.
How the citation model limits data exposure
One of the notebook's architectural features is that it will not generate content outside the boundaries of your uploaded sources. This is not merely a policy position — it is a structural consequence of how the retrieval and generation pipeline works. The model retrieves passages from your corpus, composes an answer referencing those passages, and links each sentence to its source. It will tell you when a question falls outside the material you have uploaded rather than drawing on its pre-training knowledge.
This means a shared notebook cannot be used to extract content from another user's unrelated notebook, and a collaborator with viewer access sees only the sources the notebook owner has added. The isolation boundary is the notebook itself.
Regional rollout and access
The notebook has been available globally for personal accounts since late 2023, with ongoing regional rollouts for new features. Audio overviews in non-English languages rolled out through 2024 and 2025, with some language hosts still in limited availability depending on geography. Workspace availability depends on the specific Workspace edition and whether the administrator has enabled the tool. Regional data-residency guarantees for Workspace are tied to the data-residency region selected at the organisation level, not to the notebook product specifically.
| Tier | Training use of uploads | Retention | Share scope |
|---|---|---|---|
| Personal (free) | Not used for training | Until source or notebook deleted | Account-level; shareable by invite |
| NotebookLM Plus | Not used for training | Until source or notebook deleted | Account-level; team sharing with roles |
| Workspace (standard) | Excluded under Cloud DPA | Governed by Workspace DPA | Org domain; admin controls access |
| Workspace for Education | Excluded; additional student-data terms | Governed by Workspace for Education terms | Restricted to org; admin managed |
What the tool does not promise
Transparency means stating what the data terms do not cover as clearly as what they do. The training opt-out does not apply to metadata about usage patterns — how often you use the tool, which features you access, aggregate performance telemetry. That data falls under standard Google product usage terms. The citation model keeps content grounded but does not prevent a collaborator with appropriate permissions from reading the sources you have added to a shared notebook. Sharing is intentional and role-controlled, but it is sharing.
Data and privacy questions
Straight answers to the questions researchers, enterprise buyers, and educators ask most often about how the tool handles uploaded material.
Does NotebookLM use my uploads to train Google's AI models?
No. Google states that content uploaded to the notebook by personal-account users is not used to train its foundation models. This is the default — no setting needs to be changed. Workspace customers are covered by the Cloud Data Processing Addendum, which carries the same exclusion.
How long does the tool retain uploaded sources?
Sources remain available as long as the notebook exists. Deleting a source removes it from the active index almost immediately. Deleting the notebook removes all associated content. After account closure, Google's standard data-deletion timelines apply.
Is Workspace data handling different from a personal account?
Yes. Personal accounts are governed by the Google Privacy Policy and the product's own terms. Workspace accounts operate under the Cloud Data Processing Addendum, which adds data-residency controls, admin audit logging, and the ability for administrators to restrict which users can access the tool. Workspace for Education accounts carry additional protections for student data.
Can answers leak content from one user's sources to another?
No. Each notebook is isolated to the account that owns it. The retrieval model only draws on sources inside the active notebook. It cannot access another user's notebooks, and its grounding rule means it will not generate answers from outside your uploaded corpus. The isolation boundary is structural, not just a policy.
Understand the full feature set
Privacy is just one dimension. The features page covers what the tool can do once you are confident about how it handles your material.
Explore NotebookLM featuresFurther reading on NotebookLM
Readers researching the privacy side of this tool often also look at the NotebookLM pricing page to understand tier differences, and at NotebookLM Plus for the enhanced sharing and admin controls. The Gemini and NotebookLM page explains the model architecture underlying the retrieval system. Understanding sources and uploads helps clarify what data enters the notebook in the first place. Enterprise readers frequently consult our NotebookLM guide before deploying the tool across a team.
For independent context on AI data governance, the FTC privacy and security guidance provides a regulator's framework for evaluating any AI tool's data commitments. The about page describes the editorial standards this reference applies when documenting policy claims. If you have a specific data-handling question not covered here, the contact page is the right place to send it. The help desk page covers support paths for account-level issues. The features overview and the step-by-step walkthrough complete the picture of what the tool does day-to-day.