Google states that it does not use your NotebookLM uploads, queries, or the responses you get to train its foundational models, with one exception. On a personal account, if you choose to submit feedback, human reviewers may see the related content to improve the product. Google disconnects your feedback from your account before reviewers see it. Workspace and Education uploads are not used for training even then. The detail most people miss is the feedback step, not the upload itself.
What NotebookLM is
NotebookLM is Google's research assistant that grounds its answers in sources you upload, such as documents, PDFs, and notes. Instead of answering from general training, it reads the material you give it and responds from that, with citations back to your sources. Because you feed it your own files, the natural question is what Google does with them. The answer depends on your account type and on whether you opt into feedback.
The training question, by account type
For consumer accounts, Google says your uploaded sources and chats are not used to train its models in the normal course of use. The exception is feedback: if you actively submit feedback on a response, human reviewers may read the related content to improve the service. For Workspace and Education accounts, the protection is stronger, since that data is handled under enterprise terms and is not used to train models. Two things decide your exposure: your account type and whether you submit feedback.
| Account type | Used to train models? | Human review |
|---|---|---|
| Personal, no feedback | No | No |
| Personal, you submit feedback | Not for foundational training | Reviewers may see the flagged content |
| Workspace / Education | No | Not reviewed, even with feedback |
What "not used for training" does not mean
Not training on your data is not the same as never processing or storing it. To answer your questions, NotebookLM still reads and processes your uploads, and content is retained while you use the product, subject to Google's policies and your deletion choices. Privacy here means your sources do not feed a model that other people query, not that the data never leaves your device. For genuinely sensitive material, the safer assumption is to treat any cloud tool as processing your content, and to control what you upload.
How to keep it private
- Skip the feedback submission on sensitive notebooks, since that is the path that can expose content to human reviewers.
- Prefer a Workspace or Education account for sensitive work, where enterprise terms apply.
- Upload only what the task needs, and leave out identifiers you do not need in the answer.
- Delete notebooks and sources you no longer need, rather than leaving them stored indefinitely.
Treat the feedback button as the privacy decision, not the upload. On a personal account, declining to submit feedback is what keeps a sensitive source out of human review.
The broader lesson
NotebookLM holds up on privacy for what it is, but the same questions apply to any cloud AI: who can read your data, whether it trains a shared model, and how long it is kept. A memory layer like MemX answers those by design, keeping your documents in a store you control, kept private by architecture rather than by a checkbox you have to remember. The point is not that NotebookLM is unsafe. It is that knowing the exact terms beats assuming them.
01Does Google train its AI on my NotebookLM uploads?
Google states it does not use your uploads, queries, or responses to train its foundational models in normal use. The exception is personal accounts where you submit feedback, in which case human reviewers may see the related content.
02Is NotebookLM safe for sensitive documents?
It is reasonable, but any cloud tool still processes and stores your content to function. For sensitive material, prefer a Workspace or Education account, avoid submitting feedback, and upload only what the task requires.
03What happens if I submit feedback in NotebookLM?
On a personal account, submitting feedback can let human reviewers read the related content. Google disconnects your feedback from your account before reviewers see it. That is the main path by which your content could be seen, so skip it on sensitive notebooks.
04Is NotebookLM for Workspace more private?
Yes. Workspace and Education data is handled under enterprise terms and is not used to train models, which is a stronger protection than the consumer default.
05Does not training mean my data never leaves my device?
No. NotebookLM still reads, processes, and stores your uploads to answer questions. Not training means your sources are not folded into a shared model, not that the data stays only on your device.
