NotebookLM (now Gemini Notebook): Researching Your Own Documents with AI

NotebookLM, renamed Gemini Notebook in 2026, answers from the sources you upload and cites them. How to use it in research, plus its limits and privacy.

info Quick Definition
NotebookLM, called Gemini Notebook since July 2026, is a Google tool that answers questions based on the sources you upload to a notebook, such as PDFs, documents, web pages, YouTube videos or audio files, and cites them. For a designer, it is useful for reading and synthesizing research material without mixing it with the rest of the internet.

What makes it different from a general chat?

A note on the name: on July 16, 2026 Google announced that NotebookLM is now called Gemini Notebook. According to Google, it remains a standalone product, and existing notebooks and sources are kept. This article uses NotebookLM because many people still search for that name, but the official help already says Gemini Notebook. That help also states that features, plans and limits change often.

A general chat answers from what it learned during training and what it finds on the web. NotebookLM works from your notebook, and its answers include inline citations to your sources (that is how a university guide for its community describes it). That reduces invented data but does not eliminate it: it can summarize a source badly or miss a nuance. That is why you should open the citations.

The practical difference is the starting point. If you upload ten interview transcripts, the conversation starts from those ten transcripts, not from a generic opinion about the topic.

What can you do with it in research?

The tasks that fit a product workflow best are:

  • Questions about a corpus: “What difficulties do users mention when paying with a foreign card?”, with the quote from each participant.
  • Summaries by theme: grouping passages from several interviews under one theme before an affinity mapping session.
  • Study guides for the team: a summary of an 80-page technical report with the points that need questions.
  • Interview preparation: checking which questions were already asked in earlier studies.

Besides chat answers, the tool generates formats such as audio and video summaries (Audio and Video Overviews), reports, quizzes and flashcards. They help when sharing with people who will not read the original material. Do not use them as a substitute for reading. Since the July 2026 update, Google also describes a feature that writes and runs code on your sources for data analysis; it rolls out by plan tier, so do not assume you have it.

What does Google say about privacy and limits?

According to Google’s help, your data is not used to train Gemini Notebook unless you send feedback. What changes is human review: in Google Workspace and Workspace for Education, uploads, queries and responses are not reviewed by people or used to train AI models. In personal accounts, if you send feedback, Google may review the full context of that interaction, including your queries and uploads. If you work with client data, use a company or institutional account, not a personal one.

Limits depend on the plan, and Google marks them as subject to change. On Google’s usage-limits help page, the free Standard tier allows 100 notebooks per user and 50 sources per notebook; paid plans raise those caps, up to 300 sources per notebook on the consumer Pro plan according to that page. Sharing a notebook does not change each collaborator’s source limit. Daily caps on queries and audio summaries are covered in a separate Google help article; check it before planning a large project, because we do not fix those numbers here.

Some universities add their own restrictions: for example, they forbid uploading health data, government ID numbers or data under ethics-board control. Before uploading real interviews, check whether they contain personal data and whether your agreement with participants allows it.

How does it apply to a project?

A simple flow for an interview synthesis:

  1. Anonymize the transcripts: remove names, emails, and anything that identifies the person.
  2. Upload only those transcripts to the project notebook, without mixing them with other studies.
  3. Ask for recurring themes, with the quote behind each one.
  4. Verify two or three quotes at random by opening the original source.
  5. Bring the themes to the team session and decide what they mean for the product.

This flow connects with AI interview synthesis and with transcription, which is usually the step before it.

When should you NOT use it?

  • With material you cannot upload under your contract or your participants’ consent.
  • For decisions based on a single document with no contrast against other sources.
  • When you need data that is not in your notebook: the tool works from your sources and does not replace an open web search.
  • When you want the tool to set product priorities. That is not its job.

Example

Illustrative example: a designer has twelve interviews about renewing an insurance policy. She anonymizes the transcripts, uploads them to a notebook, and asks about the reasons people abandon the renewal. She gets five themes, each with quotes. She opens three quotes at random and confirms they are faithful. She then takes the themes to her team, which finds that the main problem was not the price, as they had assumed, but the reminder email.

What she learned: the tool sped up the reading, but the conclusion came from the team’s discussion of the quotes.

How do you prepare the sources before uploading them?

The quality of the answers depends more on the sources than on the model. Before you upload material, do three things:

  • Name each file with a fixed pattern: date, type, and participant code, for example 2026-09-14_interview_P07_anonymized.txt. That lets you cite and trace without opening everything.
  • Remove what is not evidence: forwarded emails, signatures, attendee lists. Noise in a source produces summaries with noise.
  • Separate by study: one notebook per research project. Mixing studies from different clients or years makes an answer combine contexts that are not comparable.

A practical tip: upload two sources first and ask three control questions whose answers you already know. If the notebook fails on the obvious, do not keep adding material until you understand why.

How do you turn an answer into a finding?

An answer with a citation is not yet a finding. To turn it into one, write a sentence that starts with the subject (who), continues with the behavior or need observed, and ends with the quote that supports it. If you cannot write the sentence without using words that do not appear in the source, the answer is inflating the data.

Common mistakes

  • Uploading interviews with personal data without anonymizing them.
  • Believing an answer is correct because it has a citation, without opening the source.
  • Mixing studies from different years and clients in one notebook.
  • Trusting limits from a university article, or from this one, without checking Google’s current documentation.
  • Asking for a synthesis without knowing which research question you want answered.

Sources

Frequently asked questions

What is NotebookLM?
It is a Google tool that works with the sources you upload, such as PDFs, documents, or links, and answers based on them. Since July 16, 2026 Google calls it Gemini Notebook; it is the same product, and you will still see the NotebookLM name in older materials and help pages.
Does NotebookLM use my documents to train its models?
According to Google’s help, your data is not used to train Gemini Notebook unless you send feedback. In Workspace and Workspace for Education, uploads, queries and responses are not reviewed by people. In personal accounts, if you send feedback, Google may review the full context of that interaction.
How many sources can I add to a notebook?
On the free Standard tier, up to 50 sources per notebook and 100 notebooks per user; paid plans raise those caps. Google marks these limits as subject to change, so confirm them on its help page before planning a large project.
Does NotebookLM replace interview analysis?
No. It helps you find themes and quotes in your transcripts, but deciding what they mean for your product is still the team’s work.