NotebookLM Changed, Here's How to Master it Again

Google just released what might be the biggest notebook LM update since launch and they buried most of the details in a single blog post. So I spent the last week testing every single change and I can safely say that if you are still

using this tool the way you did 3 months ago, you are missing about 80% of what it can actually do. Every time you ask Notebook LM a question, there's a system that determines how much of your sources the AI can actually read during that

conversation. Up until a few weeks ago,

conversation. Up until a few weeks ago, that system had a hard limit, which meant that if you had a large notebook with 30 or 40 sources, the answers would get vague because the tool was only looking at a fraction of what you uploaded. Google just rebuilt that

uploaded. Google just rebuilt that system entirely. The amount of source

system entirely. The amount of source material Notebook LM can process in a single conversation jumped by eight times. So, if you have a notebook loaded

times. So, if you have a notebook loaded with research papers, reports, and transcripts, the AI holds all of it in context at once instead of picking and choosing which parts to reference.

Conversation memory also got six times longer, which means you can have a long complex back and forth without the AI forgetting the beginning of the chat.

Google said this update led to a 50% improvement in response quality for large source collections. And after

testing it myself, the difference is noticeable immediately. The answers are

noticeable immediately. The answers are sharper, more specific, and they pull from sources that the old version would have ignored entirely. A stronger engine means the AI can process more of your material, but what it does with that

information still depends on what you tell it to do. And that part got a major upgrade, too. Notebook LM has always had

upgrade, too. Notebook LM has always had a settings panel where you could configure how the AI responds. You could

choose response length and write a short custom instruction. The problem was that

custom instruction. The problem was that the instruction field maxed out at 500 characters, which is about two sentences, enough to say respond like a researcher, but not much else beyond that. That limit just went from 500 to

that. That limit just went from 500 to 10,000 characters, which is a 20 times increase. To show you why that matters,

increase. To show you why that matters, I'll set up a notebook and run the same question through two different configurations. First, with a basic

configurations. First, with a basic instruction in the settings, I'll type the following. act as a research

the following. act as a research analyst. The response will be organized

analyst. The response will be organized and relatively useful, but it will read like a generic summary that could have come from any AI tool. Now, I'll replace that with a detailed instruction and paste in this prompt. It's basically a

prompt structured like a job description for the AI. Instead of just asking a question, you first tell it who it is and how it should think. It forces every answer to follow a clear format, a quick

summary, a deeper explanation, and a few open questions at the end. It also tells the AI to focus on real research, label what type of sources it's using, show both sides if sources disagree, and avoid guessing or giving generic info.

So, instead of getting a normal chatbot answer, you're turning it into a serious research assistant that's structured, evidence-based, and much more reliable.

That is still well under the limit, but it changes the entire output. The AI now automatically filters facts from opinions, flags contradictions, and structures the answer exactly the way you want it. And the community has already started building on this.

There's a GitHub repository called Awesome Notebook LM Prompts where people are sharing tested configurations for everything from modern newspaper style to premium product showcases. This

10,000 character capacity turns every notebook into a unique tool allowing you to view the exact same documents from completely different angles. So, the

chat side of notebook LM is a completely different tool now, but the studio panel where you actually generate content from your sources got just as big of an overhaul. But before we get into those,

overhaul. But before we get into those, I want to quickly show you Higsfield because it's a great example of how fast AI tools are evolving right now.

Higsfield combines several leading image and video models in one platform, including Nano Banana Pro, Sora 2, VO 3.1, Cling, plus additional tools, all under a single subscription. You can

generate an image, then take that same image and turn it into a video with real camera movement and depth, all inside the same platform. And having access to this many models in one place means you can pick the right model for the job

instead of being locked into whatever one tool offers. I've left a link to Higsfield in the description if you want to try it out. And a huge thank you to Higsfield for sponsoring this video. The

audio overview has always been one of the things that sets Notebook LM apart.

You hit generate and it produces a podcast style conversation between two AI hosts discussing your material. The

problem was you only had one format, the deep dive, which was a long conversational breakdown of everything in your sources. It was good, but not every situation calls for a 15-minute podcast. Google just solved that. And

podcast. Google just solved that. And

now, instead of a single default style, you can generate audio in four different ways. Brief gives you a 1 to2 minute

ways. Brief gives you a 1 to2 minute summary. If you need to quickly

summary. If you need to quickly understand what a set of documents covers without committing to a full session, this is what you use. I've been

generating these for notebooks where I just need the highlights before deciding whether to go deeper. Critique is where the two hosts go through your sources and point out what's weak. They'll call

out gaps in the argument, missing evidence, and spots where the reasoning doesn't hold up. If you're working on a research paper or a presentation, running a critique before you finalize it catches things you'd miss just reading through it yourself. And debate

puts the two hosts on opposite sides.

They pick a topic from your sources and argue it back and forth. I found this especially useful when my sources don't agree. Because instead of reading

agree. Because instead of reading through everything and figuring out who was saying which point, the debate simply lays out both sides for you. Each

format still accepts custom instructions so you can tell it exactly what to focus on, what tone to use, and how long to run. and you can regenerate until the

run. and you can regenerate until the output matches what you need. But Google

didn't stop at audio because the Studio Panel just added an entirely new output type that none of the other tools have.

Up until now, every output in Notebook LM was either text, audio, or a visual like an infographic or slide deck. If

you needed to compare information across your sources in a structured way, you'd ask the chat to format it as a table, copy it out, and paste it into a spreadsheet yourself, which usually messes up the formatting completely.

Data tables is a new output type in the studio panel that fixes this. You tell

Notebook LM what you want to compare and it generates a structured spreadsheet style table pulled directly from your sources exportable to Google Sheets in one click. I'll click on data tables in

one click. I'll click on data tables in the studio panel and type create a comparison table of all approaches discussed in my sources including the key researchers, primary methodology, and main criticism of each approach. It

returns a clean comparison table with one row per approach and columns for each specified parameter, plus a source column indicating the reference document for that row. What used to take hours of manual copying, pasting, and

reformatting now happens in seconds with perfect accuracy. For anyone doing

perfect accuracy. For anyone doing literature reviews, competitive analysis, or cross document comparison, this cuts hours of manual work. On the

visual side, Notebook LM now gives you full creative control over how your insights are presented. First, you can choose the perfect orientation for your project. Whether that's a wide landscape

project. Whether that's a wide landscape view, a classic portrait layout, or a square format perfect for social sharing. Next, you can dial in the level

sharing. Next, you can dial in the level of detail. You can keep it concise for a

of detail. You can keep it concise for a highle summary. Go with standard or

highle summary. Go with standard or select detailed if you want to really lean into the data. But the real power lies in the custom prompt box. Instead

of leaving it to chance, you can describe exactly the aesthetic you're after. I'll type create this infographic

after. I'll type create this infographic in a clean editorial magazine style with serif typography and muted colors. And

the output comes back noticeably different from the default. The biggest

upgrade is how precisely you can describe what you want. Instead of

relying on presets, you simply write out your idea in plain language and the model follows your description closely.

That means you have direct control over the details, composition, and overall direction. And in most cases, you get

direction. And in most cases, you get something that matches your original vision much more accurately than before.

And that makes the last update even more interesting because Google just connected Notebook LM to a tool that can take all of that research and do something completely different with it.

Notebook LM has always been a closed system. You work with your sources

system. You work with your sources inside the notebook and everything stays there. That was fine for research and

there. That was fine for research and analysis. But the moment you wanted to

analysis. But the moment you wanted to create something from that research, you had to leave notebook LM and start over in another tool. But now you can use your notebook LM notebooks as a source directly inside the Gemini app. So I can

take this research notebook I've been building, open Gemini, and add the notebook as a source. Now when I ask Gemini to write something like develop key takeaways, it pulls from the curated sources I already organized in notebook

LM instead of its general training data.

You can also add notebooks to custom gems, which means you can build a gem that specializes in your specific research area and always references your verified sources. The research happens

verified sources. The research happens in notebook LM where it's grounded and cited, and you take that work further in Gemini, where you have access to Nano Banana Pro, Google VO, and everything else. The 10,000 character prompt field

else. The 10,000 character prompt field is the update I keep coming back to because that one change turns every notebook into a specialized tool built exactly for how you work. With Notebook

LM rolling out updates so quickly, the difference between power users and regular users is only going to get bigger. So, if you want to keep up with

bigger. So, if you want to keep up with these changes as they happen, go ahead and subscribe. Thanks for watching, and

and subscribe. Thanks for watching, and I'll see you in the next one.