How to Use NotebookLM: Step by Step Guide

Brian Burchanowski

TL;DR: NotebookLM, a free Google tool, serves as a personalized AI expert in the info you add, helping you research & study, answering Qs, finding patterns, and even critiquing your work.

What is NotebookLM?

NotebookLM is a new and unique offering from Google: A Large Language Model built with Gemini 1.5 that focuses exclusively on the information sources you add, serving as a personalized AI expert in your information

Recognizing that people are struggling with information overload, Google introduced NotebookLM on July 12, 2023 to help synthesize facts and ideas from multiple sources. “A key difference between NotebookLM and traditional AI chatbots is that NotebookLM lets you ‘ground’ the language model in your notes and sources.”

As your research and study assistant, NotebookLM can take notes in order to capture and collate information. NotebookLM can provide easy access to all the information you need out of the mountains of data you add. Serving as a librarian-style personal assistant, the tool can collaborate with you, helping you conduct research, write a paper, and develop your project.

What Sources Can You Upload to NotebookLM?

Initially, you could upload only Google Docs, PDFs, and text files. In June 2024, Google added features, including the capacity to upload more sources and the ability to upload Google Slides and website URLs as sources, as well as YouTube URLs and audio files. The current limit is 50 sources with a maximum of 500,000 words each. Specifically, with YouTube URLs, the videos must be public and have an accessible transcript.

A screenshot of a computer

Description automatically generated

How Much Does NotebookLM Cost?

NotebookLM is still in the early testing phase, so there are currently no fees for using it. It’s an experimental product, built by a small team in Google Labs that has two goals in mind:

  1. Roll out this technology responsibly
  2. “Build a product with our users: We’ll be talking to people and communities often to learn about what’s working well and where the gaps are, with the intent of making NotebookLM a truly useful product.”

You can provide feedback within NotebookLM, via Google Doc, and on the NotebookLM Discord server.

How Does NotebookLM Organize Info & Stay Up to Date?

Navigating through the sample Notebook is more reminiscent of a binder than a notebook, since sources are organized into tabs you could click on to expand each section. If you create multiple notebooks, each remains separate, so NotebookLM can’t access information from various notebooks simultaneously.

Sources are automatically kept up to date. When you upload a Google Doc, Slide, or a local PDF or text file from your computer, the app makes a copy of the original file. NotebookLM will notify you if the original Doc or Slide has changed, at which point you have the option to update the source to refresh its content.

How Does NotebookLM Answer Questions?

NotebookLM uses Gemini 1.5’s multimodal capabilities to assess and make interesting connections between the sources you’ve added. Similar to Gemini in Google Workspace, users can get assistance with creative tasks like brainstorming, in addition to responses to factual inquiries.

The Notebook Guide not only automatically summarizes uploaded material, but also proposes insightful questions and offers options for generating content in various formats like timelines or study guides. You can directly ask questions, and responses are generated solely from uploaded sources. You’ll receive answers with inline citations. The citations show you the most relevant original quotes from your sources, showing the exact source, not just the answer.

How Do You Take & Use Notes with NotebookLM?

There are two kinds of “notes” in NotebookLM.

  1. Saved Responses that you have captured, either chat interactions or quotes from your sources
    1. Saved Responses often have citations, so you can easily jump back to their original location in a source
    2.  Saved responses are “view only” and cannot be edited
  2. Written Notes that you have composed yourself
    1. Written notes can be edited

Google describes “Curate and Create” as one of NotebookLM’s most powerful workflows. Highlighting notes will result in suggested actions, including: 

  • Combine a group of notes into a single note
  • Summarize the contents of the notes
  • Organize them into thematic categories
  • Create a study guide
  • Create custom formats like blog posts, marketing plans, or email newsletters

You can use a collection of notes you’ve written yourself as a kind of source: just select all your notes and ask NotebookLM a question about their contents. NotebookLM’s answers will be grounded in the information you’ve written.

Can NotebookLM Hallucinate?

Technically yes, NotebookLM can hallucinate, like any AI. However, the risk is relatively lower. As written in a Google blog, “While NotebookLM’s source-grounding does seem to reduce the risk of model ‘hallucinations,’ it’s always important to fact-check the AI’s responses against your original source material.”

Additionally, The Verge reports, “Raiza Martin, senior product manager at Google Labs, told reporters that NotebookLM ‘is a closed system.’ It will not do any web searches beyond reading the website content users add.”

NotebookLM can create “Audio Overviews,” which are similar to podcasts, with hosts discussing your content in conversational tones and using metaphors. Google noted that the audio conversations are still in beta and there may be some inaccuracies in the conversations.

How to Get Started with NotebookLM

To get started, visit https://notebooklm.google.com/. You can select, “Introduction to NotebookLM,” to browse and experiment with an existing Notebook about NotebookLM.

A screenshot of a computer

Description automatically generated

Select, “Create new,” and you’ll immediately be prompted to upload sources.

A screenshot of a computer

Description automatically generated

Creating a Cybersecurity Guide with NotebookLM

For this example, cybersecurity was selected as the subject matter to exhibit the process. Cybersecurity is a suitable topic due to the availability of relevant, public, and non-copyrighted resources. Additionally, cybersecurity is a topic of interest for Wursta’s audience of information technology professionals.

 To start, upload all of your documentation. Numerous websites and PDF files from cybersecurity regulatory bodies and organizations were uploaded, all listed below.

By default, NotebookLM will consider all of your uploaded resources when responding to questions. However, you can also select one source and receive answers based solely on information in that source, which is known as a “focus state.” Just select the sources you want to focus on in the left-hand sidebar and enter your prompts to conduct your research and investigate your sources. The “focus state” will be displayed in the left-hand side of the chatbox: it will state “# sources” or “selected notes” when applicable.

A screenshot of a computer

Description automatically generated

Using Gemini 1.5’s multimodal capabilities to assess and make interesting connections between the added cybersecurity sources, NotebookLM automatically suggested the following questions:

  • How does each standard define the scope of its requirements?
  • What are the key similarities and differences between these frameworks?

The response to the latter was very long.

A screenshot of a computer

Description automatically generated

The next  prompt entered was, “Summarize those similarities and differences – giving me only the highest level bullet for each category,” (Similar to Google web search, NotebookLM understands typos.)

After receiving the summary response, you can click “Save to Note,” however, in this case, you’re unable to edit the note. “Saved responses are view only.”

This can be easily resolved by copying and pasting the text to a written note, which offers basic text editing capabilities.

NotebookLM delivered the following summarized response to the initial query, “What are the key similarities and differences between these frameworks?”

Similarities: all frameworks: As shown in the list of sources below, the frameworks include NIST CSF, PCI SSC, CMMC, and more.

  • Are risk-focused
  • Use layered security
  • Are flexible and adaptable
  • Support continuous improvement

Differences: each framework has a different:

  • Scope and area of focus
  • Level of prescriptiveness
  • Target audience
  • Approach to assessment and certification

Asking to summarize further yet to one bullet yielded:

Similarities

  • All frameworks aim to improve an organization’s security posture.

Differences

  • Frameworks differ in scope, prescriptiveness, target audience, and assessment/certification processes.

Responses also have proper formatting, for example, larger font and initial caps for headers.

Save Important Facts and Ideas 

With the “Noteboard,” you can pin things in order to save interesting findings.

A screenshot of a chat

Description automatically generated

Website Sources Aren’t Scanned?

The use of websites as a source is a bit flawed. NotebookLM suggested the prompt, “Evaluate Check Point Research’s use of different media (podcasts, reports, news articles) to disseminate information.”

Mousing over the cited sources in the response reveals blog titles, including the amusing “January 16, 2023, AI Can Write Malware Now. Are We Doomed?” However, it’s not possible to click a link to view this blog directly. 

A screenshot of a computer

Description automatically generated

Prompting, “I’m interested in this blog you pulled up, AI Can Write Malware Now. Are We Doomed? can you provide a direct link to it?” yielded a response equating to, ‘Nope.’

A screenshot of a chat

Description automatically generated

“The sources provided only list the titles and dates of Check Point Research’s publications, and they do not provide direct links to the content.”

Interesting. This begs the question, is NotebookLM pulling data from within the site, including the body text of blogs, or merely page titles?

The next prompt entered was, “Are you scanning and indexing the content of those blogs or only reviewing the titles?”

The response was a bit unclear. The blog in question includes the subheading, “[AI] can write an entire attack chain: phishing emails, macros, reverse shells, you name it.”

To further investigate this issue, “is it possible for ai to write an entire attack chain?” was asked.

NotebookLM replied that there is no mention of whether AI can, serving as evidence that NotebookLM appears to not scan the contents of websites listed as sources.

A screenshot of a chat

Description automatically generated

However, the ability to submit a website as a “source” should lead to NotebookLM scanning the interior of the site, and not merely page titles. At the very least, this unexpected behavior should be clearly communicated to users. That was submitted as feedback.

NotebookLM Can Critique Your Work

Once you have text in a written note, you can select the note and choose “critique,” and NotebookLM will offer constructive criticism about the argument you’re making.

One critique of NotebookLM is that you can’t adjust the width of the lefthand column, so it’s difficult to read the titles of each section without mousing over. It’s possible to rename sources, but the max that is displayed is about 20 characters, which is still rather short. 

Give NotebookLM a Try & Roll Out Gemini for Workspace

While NotebookLM is a generative AI tool, its grounding in data that you select and upload makes it unique. It’s fascinating to experiment with and we recommend trying it as part of your organization-wide use of Gemini.

And if you haven’t yet rolled out Gemini for Workspace, contact Wursta. Our three-step, 3E Framework is designed to guide you through every step of the process, from initial assessment of Gemini to full-scale implementation, empowering your team to adapt and thrive in the era of intelligent automation.

Sources for Wursta’s Sample Cybersecurity Notebook

PDFs

Websites

See Similar Articles