Does NotebookLM train on your data?
No. Google states that NotebookLM does not use your uploaded sources, queries, or the content you create to train its AI models. Your data is not shared, and personal data uploaded to NotebookLM is not used to train models offered to the general public. Workspace and enterprise editions carry additional data protections.
Why — the first-principles explanation
There is a difference between the model and your session. NotebookLM runs on Google's pre-trained Gemini model, which was trained beforehand on other data. When you upload documents, they are fed to that already-trained model at the moment you ask a question, they are not folded back into the model's training.
Google's stated policy is that it does not train its models on the personal data you upload to NotebookLM. This is the key privacy promise: your PDF or notes stay within your notebook to answer your questions, rather than becoming part of a model that other people query.
There is one normal caveat. If you submit feedback (like thumbs-up/down) or if human review is needed for abuse or safety, trained reviewers may see limited data, a standard practice across AI products. But that is review, not training, and Google's paid Workspace/enterprise tiers add further contractual data protections. Always check the current in-product privacy notice, since policies can evolve.
An example that makes it click
Imagine a tutor who has already finished all their schooling. You hand them your essay to critique. They read it, give feedback, and hand it back, but reading your essay doesn't change what they learned in school; it doesn't get added to their permanent knowledge.
NotebookLM is like that tutor. It uses what it learned before to help with your document, but your document isn't absorbed into its 'education' for the next student.
Key facts
- Google states NotebookLM does not use your personal uploaded data to train models for the general public.
- Google's help center states your data will not be shared by NotebookLM.
- Your sources are used to answer your questions, not folded into model training.
- Human reviewers may see limited data only for feedback or safety review, which is not training.
- Workspace and enterprise editions include additional contractual data protections.
Upload your sources and get grounded answers and audio overviews.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
Does NotebookLM train on your data? According to Google, no. Your uploaded documents, your questions, and the content you create are not used to train Google's AI models for the general public. Here's why that makes sense technically. NotebookLM runs on Gemini, a model that was already trained ahead of time on other data. When you upload a PDF, it's fed to that finished model just to answer your questions in the moment. It doesn't get baked back into the model's brain. Google's help center is direct about it: your data will not be shared by NotebookLM. There's one normal caveat, if you send feedback or if something needs a safety review, a trained human might see limited data, but that's review, not training. And the paid Workspace and enterprise versions add extra data protections in the contract. Bottom line: your sources stay yours, used to help you, not to teach a model other people use. Still, it's worth reading the in-app privacy notice, since policies can change.
What authoritative sources say
People also ask
Can Google employees read my documents?
Only limited data may be seen by trained reviewers for feedback or safety review; it is not routinely read or used for training.
Is my uploaded PDF added to Gemini's training?
No. Google states personal data uploaded to NotebookLM is not used to train its publicly available models.
Is the paid version more private?
Workspace and enterprise editions add contractual data protections on top of the standard privacy commitments.
Where can I confirm the current policy?
Check the in-product privacy notice and Google's NotebookLM help pages, since policies can be updated over time.