Notebook LM is a powerful research intelligence system, not just a basic chatbot, capable of autonomous sourcing, validation, and generative workflows that transform raw data into actionable insights and content.
Mind Map
Click to expand
Click to explore the full interactive mind map • Zoom, pan, and navigate
Right now, the average user treats
Notebook LM like the usual basic
chatbot. They upload a file or two, ask
a question, and maybe click around
hoping to find something useful. But
that is a massive underutilization of
the tool because Notebook LM isn't just
a chatbot. It is a complete research
intelligence system. We are talking
about autonomous sourcing that finds
data for you, validation methods that
ensure your research is reliable, and
even generative workflows that turn that
data into slide decks and infographics
instantly. So, in this video, I'm going
to walk you through the complete
advanced workflow to get you using this
tool better than 99% of people. And by
the end of this, you'll know exactly how
to use Notebook LM the way it is
actually meant to be used. Let's get
started. First, I'll head over to the
Notebook LM website. And on the
homepage, you'll see your notebook list
if you've created any before or an empty
state if this is your first time. So, we
can create a new notebook here. But
before we do, here is a critical detail
that often gets overlooked. It's that
notebooks should be topic specific.
Don't create a notebook called research
or general notes. create focus notebooks
like competitive analysis Q12025
or AI video generation research. And
that might seem insignificant, but it
actually matters because notebook LM
performs better when sources are related
and focused on a single topic or
project. I'll go ahead and create a new
notebook here by pressing this plus
icon. And this immediately takes you to
the source upload screen. This is where
the quality of your output gets decided.
The common mistake here is simply
uploading one or two files without
thinking strategically about what
sources you actually need. Instead, take
a step back. Think about the complete
information landscape for your topic.
What formats does your information exist
in? PDFs, YouTube videos, websites,
Google Docs? The power of Notebook LM
comes from combining multiple formats
and sources, creating a web of
information rather than just a stack of
isolated documents. And later in this
video, I will show you exactly how to
professionally execute that multiiformat
strategy. So, make sure you stick around
for that. For this example, I'm going to
create a research notebook on AI
alignment. Let's hit escape, and I'll go
right up here and title this AI
alignment research. Now, here's a
feature that was added quite recently,
which is the deep research source
discovery. Previously, there was only
fast research, which was good but
limited. But now, you can access a much
more powerful agent by clicking this
drop down arrow and selecting deep
research. So, I'm going to select that
and write AI alignment and safety
challenges. Hit submit. And here's what
happens. Notebook LM launches an agentic
AI tool that autonomously researches
your topic. It doesn't just do keyword
matching like a basic search. It
actually analyzes the topic, finds
sources, evaluates them, adapts its
search strategy to fill gaps, and
generates a comprehensive research
report, which means deep research will
be able to discover around 50 sources
related to your topic. It generates a
detailed research report synthesizing
those sources. Then it selects the most
relevant sources and imports them
directly into your notebook. What you
get is both a curated research report
and high-quality sources already loaded
and ready to work with. And that matters
a lot because most people spend hours
manually searching for sources,
evaluating quality, and uploading them
one by one. Deep Research does that work
in minutes and often find sources you
wouldn't have discovered manually. All
right, it's finished. I now have a
curated list of citations imported
automatically, plus a full research
report that's also added as a source.
Scrolling down, you'll notice some
sources might fail to import if they're
behind pay walls. There's a remove all
failed sources button that cleans those
up in one click instead of deleting them
individually, which I'm going to do
right now. Now I have a strong
foundation of sources to work with.
Before we start asking questions or
generating content, here's a step that
gets skipped nine times out of 10, which
is source validation. Notebook LM is
extremely good at reducing
hallucinations because it grounds
everything in your sources. But that
only works if your sources are reliable
and current. If your sources are
outdated, biased toward one perspective,
or mixing primary research with opinion
pieces, Notebook LM will give you
answers based on flawed information
without distinguishing between them. So
here's the validation framework I use
for every single notebook. go to the
chat interface in the center of the
screen. Before asking any topic
questions, I run through these checks.
First, I ask, create a table showing
each source with its publication date,
author credentials, and whether it's a
primary source, secondary analysis, or
opinion piece. This gives me a clear
view of what I'm actually working with.
If I see that most of my sources are
from 2020 or earlier on a fast-moving
topic like AI, I know I need newer
material. If everything is opinion
pieces with no primary research, that's
a problem. Let me ask that now. Notebook
LM is generating a table analyzing all
of the sources. I can immediately see
the spread, when these were published,
who wrote them, and what type of source
each one is. In this case, I'm seeing a
good mix of recent academic papers,
industry reports, and technical
documentation. Most of them are pretty
recent, which is what I want for current
AI alignment research. Second, I ask
which of these sources are most
frequently cited or referenced by other
sources in this notebook. This shows me
which sources are foundational to the
topic versus which ones are peripheral.
The highly cited sources are usually the
ones I should prioritize when I'm
filtering sources later. And third, I
ask, summarize the primary perspective
or bias of the top five most substantial
sources. This tells me whether I'm
looking at this topic from multiple
angles or whether all my sources share
the same viewpoint. For controversial or
evolving topics, you want diverse
perspectives. For technical
documentation, perspective matters less.
These three checks take about 5 minutes
total, but they give me a complete
picture of my source quality before I
build my entire workflow on top of it.
With our sources validated, the next
critical step is configuration. This is
something the vast majority of users
ignore, but it dramatically improves
response quality. In the top right
corner, click right here. This opens
settings that control how Notebook LM
responds to you. First, set your
conversational goal. You have three
options. Default for general research,
learning guide for educational content,
or custom for specific use cases. For
this research notebook, I'm choosing
custom, and I'll define the role as
research analyst focused on AI safety
and alignment debates. This tells
notebook LM to frame all responses from
that perspective instead of giving
generic answers. Next, choose response
length. You have default, longer, or
shorter. For research work, I typically
choose longer because I want detailed
analysis, not brief summaries. Click
save. These settings now apply to every
chat in this notebook. You set them once
and forget about them, but they shape
every interaction from this point
forward. The majority of the people use
notebooks in default mode and wonder why
responses feel generic. Configured
settings give you targeted rosp specific
answers optimized for your exact use
case. Now let's look at how to work with
sources strategically instead of just
accepting all of the sources for every
query. On the left side you'll see your
source list with these checkboxes next
to each file. And a very common mistake
that people make is that they leave
everything checked all the time. When
you ask a question with all of the
sources selected, Notebook LM tries to
synthesize an answer from every single
document. This dilutes your results. It
forces the AI to generalize, giving you
a vague surface level summary instead of
a deep answer. So let's say I want to
focus specifically on existential risk.
If I leave the mechanistic
interpretability checked, I am confusing
the model by forcing it to look at
conflicting topics. So I'm going to
uncheck everything. Then I will go
through and select only the three
technical papers that contain the actual
code logic. Now effectively the other
documents do not exist to the AI. It can
only see what is checked. When I ask,
how do these agents handle memory
management? Notebook LM creates the
answer exclusively from those three
technical papers. The answer comes out
sharper, more technical, and completely
free of irrelevant information. This
gives you surgical control over your
research. You can keep one massive
master notebook with 50 sources, but by
toggling these check boxes, you can
instantly turn it into a focused subnote
for any specific query. All right, now
let's generate some content from our
sources. We'll start with an audio
overview, which is one of Notebook LM's
signature features. On the right side,
you'll see the studio panel. Click on
audio overview. Now, don't just click
generate yet. Most people just blindly
hit generate and accept whatever random
conversation the AI spits out. If you
want a result, you can actually use for
work. You need to take control of the
conversation first. In the instruction
input box below is where you tell
notebook LM exactly what to focus on,
what tone to use, and how long the
overview should be. For this research
notebook, I don't need a balanced
overview of all of the sources covering
every aspect of AI alignment. I need the
podcast to focus specifically on the key
debates and disagreements we identified
earlier. So, I'll write, "Focus
exclusively on the main disagreements
between AI safety researchers regarding
alignment approaches. Explain each
perspective clearly and keep the
discussion under 15 minutes. Use
accessible language, avoiding
unnecessary jargon. Above the
instruction box, you have two critical
settings, which are format and length.
For format, you aren't limited to the
standard deep dive option. You can
switch to brief if you need a quick
summary, or select critique, which
essentially turns the AI into a strict
editor that reviews your material for
gaps and weaknesses. But since our
prompt is specifically asking to uncover
disagreements, I'm actually going to
switch this to debate. This instructs
the host to actively illuminate
conflicting perspectives rather than
just having a friendly chat. For length,
you can choose short or default. I'll
keep this on default, which usually
gives us a solid 10-minute discussion,
perfect for digging into the details
without broadening the topic too much.
Now, click generate. Notebook LM will
take a few minutes to create a custom
podcast with two AI hosts discussing
your sources based on those specific
instructions. The difference between
default audio and customized audio is
massive. The default version covers
everything equally. The customized
version becomes a targeted research
brief focused on exactly what you need
to understand. And here's a pro tip. Do
not hesitate to regenerate. Think of the
first pass as a rough draft. If it came
out too technical, regenerate it with
instructions to simplify the language.
If it wasted time on background history,
tell it to cut the intro and focus only
on current debates. Most people generate
once and just accept whatever they get.
But the top users iterate on these
instructions until the output matches
their specific research goals perfectly.
While that audio overview is generating,
let's create visual content using
another brand new feature on Notebook
LM, which is the infographic generation
powered by Nano Banana Pro, which is
Google's advanced image generation
model. To access that, click infographic
in the studio panel. You'll see three
main settings to configure here. First
is orientation, where you can choose
landscape, portrait, or square. Next is
level of detail, which ranges from
concise to detailed. And finally, you
have the custom instruction field. For
most use cases, I recommend standard
detail level and landscape orientation.
The detailed option can introduce minor
text errors with complex topics, and
concise sometimes oversimplifies. In the
instruction field, I'll write, "Create a
professional infographic mapping the
different AI alignment approaches and
the key researchers associated with each
approach. Use clean design with blue and
gray color scheme and hit generate."
This will take a couple of minutes and
what comes back is a fully designed
infographic pulling information directly
from your sources, including talking
charts, diagrams, text hierarchies,
visual layouts, everything you'd
normally need a designer to create. The
quality is legitimately publication
ready. Minor spelling errors can appear
in detailed mode with complex topics,
but standard mode is consistently
accurate. All right, here's the result.
This is a clean, well-designed visual
representation of AI alignment
approaches with key researchers mapped
to different strategies. The design is
professional. The information is
accurate and cited from my sources, and
this would have taken hours to create
manually. You can also regenerate this
with different instructions if you want
to adjust the style or focus. Next,
let's create a presentation deck, which
is the other new Nano Banana Pro
feature. In the studio panel, click
slide deck. You'll see two deck types:
detailed deck, which creates
comprehensive slides with full text
suitable for sending as a standalone
document, or presenter slides, which
creates clean visual slides with minimal
text designed to support you while
speaking. For most presentations,
presenter slides is better because it
keeps slides visual and text minimal.
For length, you have two main choices.
Short for a 10 slide summary or default
for a full 15 to 20 slide deck. I want
just the key points, so I'm going to
choose short. In the instruction field,
I'll write create a presentation
explaining the three main schools of
thought in AI alignment for a technical
audience. Focus on key differences and
trade-offs. Click generate. This will
take a few minutes to create a fully
designed slide deck. While it's
generating, let me explain why this is
powerful. Most people spend hours
building presentations from research.
They read through sources, extract key
points, design slides, find or create
visuals, and structure the narrative.
Notebook LM does all of that
automatically. It pulls information from
your sources, structures it logically,
designs professional slides, and creates
supporting visuals. And just like audio
overviews, you can regenerate with
different instructions if the first
version isn't quite right. All right,
the deck is ready. Let's take a look.
This is a clean, professionally designed
presentation. Each slide has a clear
visual hierarchy supporting graphics and
text pulled directly from my sources
with proper structure. Slide one
introduces the topic. Slide two breaks
down the three main approaches. Each
slide explores one approach in detail
with visuals that illustrate the key
concepts. This is presentation ready
output that would normally take several
hours to build manually generated in
minutes from your sources. The audio
overview we generated earlier should be
ready now. So, let's open it. You'll see
a standard podcast player with two AI
hosts discussing AI alignment based on
our custom instructions. Let me play a
bit of it.
>> Welcome to the debate. We're diving into
what I think is probably the most
consequential question of our time. How
do we make sure that these incredibly
powerful AI systems we're building are,
you know, fundamentally aligned with
human values.
>> Audio is great for understanding the big
picture, but for precision work, we need
the chat interface. In the center panel,
you can ask any question about your
sources. The key is asking precise
questions instead of vague ones. Instead
of asking, "What does this say about AI
alignment?" asks, "Compare the three
main technical approaches to AI
alignment and explain the key trade-off
each approach makes. That specific
question gets you a structured, useful
answer. You'll also notice little
numbers scattered through the text.
Those are citations. When you click one,
it highlights the exact passage in the
original document, letting you verify
the accuracy of the text instantly. But
if you need something more engaging than
just audio, there is the video overview.
This just got a major upgrade with
custom visual styles. In the studio
panel, click video overview. This
creates a narrated explainer video with
AI generated visuals based on your
sources. It's similar to audio overview
but with slideshow style visuals that
illustrate the concepts as they're
explained. You'll see two content
options. Explainer, which creates a
comprehensive overview connecting
concepts from your sources, or brief,
which gives you a quick bite-sized
summary of core ideas. For most use
cases, explainer is better because it
provides depth and proper context. Below
is an option to choose custom visual
styles powered by Nano Banana Pro. You
can choose auto select to let notebook
LM pick a style from their preset
library. Or you can choose custom and
describe your own visual aesthetic. Let
me try custom. I'll write clean, modern
design with blue and white color scheme,
minimalist graphics, and professional
typography. You can also guide what the
AI host should focus on in the
instruction field, similar to audio
overviews. Click generate. This takes a
few minutes to create the full video
with narration, visuals, and
transitions. All right, it's finished
processing. Let's play a quick clip to
see how it handled our custom design
request. You know, this isn't just a
technical puzzle. It's a whole series of
really deep debates about the very
nature of these artificial minds. See,
to make an AI safe, you first have to
understand it. That seems obvious,
right? But that opens up this truly
fascinating question. We can see what an
AI does, but what's actually happening
on the inside? And here's the core of
the problem. Our most powerful AI models
are basically black boxes.
>> And look at that. It didn't just grab
random stock footage. It actually
followed my prompt for a clean blue and
white color scheme with minimalist
graphics. The narration is synced
perfectly with the visuals and the
structure follows the logical flow of
our source documents. This is perfect
for creating educational content,
presentation materials, or sharable
explanations of complex research. All
right, let's wrap up the studio panel by
looking at the remaining tools, which
are reports, flashcards, quiz, and mind
maps. These are all found in the studio
panel on the right, and each serves a
specific organizational purpose. Let's
start with reports. Click reports and
you'll see several options here. The
first one which we're going to look at
is the briefing dock. This creates
several pages of executive summary of
your entire knowledge base featuring key
insights and quotes from your sources.
I'll click to generate one now. And
here's the result. This is a clean,
professionally structured document
summarizing the key findings from all of
the sources. I can export this to Google
Docs, edit it if needed, and use it as a
foundation for reports or presentations.
But here's the feature a lot of people
miss. You aren't limited to these
defaults. You can click create your own
to specify the exact structure, style,
and tone you want. Let's try that. I'll
write create a technical white paper
analyzing the three main approaches to
AI alignment written for researchers
include methodology comparison and
future research directions. Hit generate
and look at this result. Unlike the
generic briefing doc, this is highly
technical. It actually followed my
structure. It gave me the specific
methodology comparison and the future
direction section I asked for. This
essentially did 90% of the drafting work
in seconds. Next, you have flashcards
and quiz sections. Flash cards generate
quick Q and A pairs for memorization.
While the quiz tool builds a full
interactive test. The value here is that
they pull directly from your sources. So
you aren't testing yourself on general
knowledge. You are testing yourself on
the specific data you just uploaded. And
finally, there is the mind map. If you
click this notebook LM generates an
interactive diagram showing how the key
concepts in your sources actually
connect to each other. You can click any
node to expand it into subtopics or
click it again to trigger a detailed
chat response about that specific idea.
This is massive for visual learners
because it helps you spot connections
between files that you would definitely
miss just by reading them linearly. And
that is the key takeaway here. It is a
mistake to limit yourself to just the
chat and audio overview. These
organizational tools are what actually
transform raw information into a
structured knowledge system. Now, to
bring this full circle, I want to
deliver on that promise I made at the
start of the video. We need to talk
about source strategy, specifically how
to mix different formats to create a
truly comprehensive research system. The
vast majority of users upload one type
of source. Maybe they add five PDFs or
maybe they add three YouTube videos, but
they don't think strategically about
combining formats. Here's what you
should do. Notebook LM accepts PDFs,
websites, YouTube videos, audio files,
Google Docs, and plain text. The power
comes from mixing these formats to cover
your topic from multiple angles. For
example, in this notebook, I can layer
YouTube lectures for accessible
explanations on top of company blog
posts for industry perspective and even
add podcast transcripts for
conversational insights. This creates a
360 degree view of the topic that you
just can't get from a single file type.
Let me add a YouTube video to
demonstrate. Click add source. Click
YouTube and paste a video URL. I'm
adding a lecture on AI alignment from a
recent conference. Notebook LM pulls the
transcript and adds it as a source. Now
I can ask questions that synthesize
across formats. Compare the technical
approaches discussed in the research
papers with the practical concerns
raised in the YouTube lecture. Notebook
LM will analyze both the written
research and the video transcript and
create a synthesis you couldn't get by
analyzing each format separately. This
multiiformat approach is especially
powerful because different formats offer
different value. Academic papers give
you rigor. Videos give you accessible
explanations. Blog posts give you
industry context. Podcasts give you
conversational insights. The typical
user stays within one format while
advanced users strategically mix every
format to build comprehensive knowledge
bases. So, at this point, you've seen
the complete workflow for using Notebook
LM, the way research professionals
actually use it. We started with deep
research to automatically build a
comprehensive source base. We validated
those sources to ensure quality and
identified gaps. We configured notebook
settings for targeted responses. We used
source filtering for focused analysis.
We generated custom audio overviews,
professional infographics, and
presentation ready slide decks. We used
mind maps for active learning. And we
built a living research system using
multi-format source mixing. The
difference between someone who uses
Notebook LM as an amateur and someone
who uses it at a professional level
isn't just knowing these features exist.
It's following the complete workflow
from source discovery through
validation, configuration, content
generation, and organization. If you
found this video valuable, you could
click right here to check out another
video I posted. It's a master class on
using Gemini 3.0 Pro at an elite level.
You'll see that these strategies like
source validation and structured
prompting don't just work in notebooks.
They are the secret to getting the most
out of Google's flagship AI as well.
Thank you so much for watching and I'll
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
