Leon Furze

Google’s NotebookLM has been out for a while, but some recent improvements have made it a much more interesting and potentially useful tool. In this post, I’ll be running through how the Gemini-based application works, and why you might choose to use it over other chatbots.

What is NotebookLM?

Google NotebookLM is an ‘experimental’ product and part of Google’s suite of language model based applications. If you haven’t stumbled across it, don’t feel bad: Google’s range of AI-based apps is frankly a nightmare to navigate, from the obtuse and ever-changing “AI test kitchen” to its developer focused (but still sandbox-y) AI studio, to standalone products like NotebookLM.

The Gemini large language model has also been incorporated into Gmail, Google Calendar, and Workspace apps like Slides, Docs, Drive, and Sheets: though again you possibly haven’t noticed even if you use these apps all the time.

NotebookLM essentially functions as a research assistant, helping users to quickly understand, synthesise, and expand upon their source materials.

Key features of NotebookLM:

1. Source Integration: Users can upload various document types including Google Docs, Google Slides, PDFs, and website URLs. The system then draws on these sources for all subsequent chats within that “notebook”.
2. AI-Powered Analysis: NotebookLM creates overviews of uploaded documents, summaries, key topics, and suggested questions to explore.
3. Intelligent Q&A: Users can ask questions about their sources, and NotebookLM provides answers with inline citations, linking directly to relevant sections of the original documents.
4. Note-Taking and Organisation: The tool allows users to create, save, and organise notes. It can transform selected notes into outlines, table of contents, study guides, or summaries.
5. Collaboration Features: Notebooks can be shared with others, enabling collaborative research and learning.
6. Suggested Actions: NotebookLM offers a range of actions to help users process information, such as combining notes, critiquing arguments, and suggesting related ideas.
7. Image Understanding: The system can analyse and reference images embedded in documents, adding another layer of comprehension.
8. Audio Guide: NotebookLM can generate short, engaging “podcast style” audio from sources.

Uploading sources

Uploading sources is straightforward, with a drag-and-drop interface and the ability to connect to Google Drive, add links, or copy/paster text directly. The source limit is around 200,000 words, meaning you can upload a significant amount of data.

Working with source material

In this example, I’ve uploaded a few of my blog posts as PDFs. These then appear in the sidebar and you can toggle them on and off when interacting with the application.

Once they’re loaded and processed, NotebookLM automatically offers some suggestions for interaction.

You can click these options, or type your own requests. In my experiments I have found that NotebookLM is good at summarisation, synthesis of the source documents, and simple Q&A. For more complex queries, it does sometimes conflate multiple documents. NotebookLM provides citations which allow you to manually check on the sources it is pulling from.

It does a good job of creating outlines, and the clickable citations make it quite useful for creating a sort of interactive table of contents across multiple documents.

Notebook Guide

The “Notebook Guide” is a relatively new feature which makes suggestions for working with the source material, and includes the new “Audio Overview” feature. It comes pre-loaded with options to create a FAQ, Study Guide, Table of Contents, Timeline, and Briefing Doc, and provides a brief summary of all the source materials.

These new documents are automatically added to the “saved notes” section of the notebook as individual items.

Probably the most impressive feature of the whole “Notebook Guide” area is the audio overview, which turns your source material into a podcast between two very lifelike AI-generated speakers. The “deep dive” podcast is an interesting and engaging addition: I’ve made a bunch of these on different source materials, and have found that the application creates unique and fairly insightful podcasts ranging from 4-15 minutes in length.

They’re a bit random, and you currently don’t have any control over the output. They are also only available in American English for now. But the results are impressive, and perhaps even a little scary: I immediately forgot I was listening to AI-generated audio because the voices and the content is so human and believable.

Here’s the full audio based on this notebook of my blog posts:

Who is NotebookLM for?

This product is clearly targeted at students and educators: the preset functions to create things like study guides and FAQs is a giveaway. The large token window of 200k words means that multiple papers, articles, structured notes, course syllabi, and other documents can be loaded up and made interactive.

You could, of course, do most of this with ChatGPT or Claude, or even just Gemini itself. But Google seems to be making a bet on smaller, more niche LM-based applications that do one job and do it well.

This is an approach I believe in myself. The issue with large, general purpose applications like ChatGPT is that they don’t come with an instruction manual and they can be pretty difficult to use well if you don’t spend lots of time with them.

NotebookLM could be an accessible, easy-to-use tool for students, and although it’s currently only in development, if it does mature it will be an application worth spending some time with.

Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch: