So What? What’s new with NotebookLM

So What? Marketing Analytics and Insights Live

airs every Thursday at 1 pm EST.

You can watch on YouTube Live. Be sure to subscribe and follow so you never miss an episode!

This episode is your NotebookLM masterclass, revealing how to build a private, trusted knowledge base for qualitative data analysis and content creation. Discover the new features—like deep research, customizable system prompts, and infographics—that leverage Gemini 3’s power while ensuring compliance and minimizing AI hallucinations.

Watch the video here:

So What? What’s New with NotebookLM

Watch this video on YouTube

Can’t see anything? Watch it on YouTube here.

In this episode you’ll learn:

What’s new with NotebookLM
Key features you should be using
Use cases for NotebookLM

Transcript:

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Katie Robbert – 00:00

Well, hey everyone. Happy Thursday. Welcome to “So What? The Marketing Analytics and Insights” live show. I am Katie, joined by Chris and John. Howdy, fellows.

John Wall – 00:42

Out of orientation today.

Katie Robbert – 00:44

B for effort.

John Wall – 00:47

Wait, I can do it in the new vertical one.

Christopher Penn – 00:50

Oh, oh, no, no.

Katie Robbert – 00:53

This week we’re talking about what’s new with NotebookLM. I’m actually excited because this is a tool that I’ve actually been really heavily using as well. So for once—I mean, I don’t know, we’ll see—but I was going to say for once, maybe it won’t be an episode where John and I just sit there and go, “Huh?” and Chris explains all the technical stuff. This is a tool I actually know pretty well at this point, so I’m pretty excited about that.

We’ve been able to get some questions from our free Slack community, Analytics for Marketers. It’s free to join if you want—the link is on the bottom—and hopefully we’ll be able to get to some of those questions because they brought up some good points. So, Chris, where the heck would you like to start?

Christopher Penn – 01:38

So let’s start really with a straightforward, “What is this thing?” I know we’ve covered it in the past, so we won’t belabor the point, but what is NotebookLM?

NotebookLM is a Google product. It is available in free and paid versions, and there are three versions. There is the free version, which allows you the basic functionality, allows you to upload up to 50 sources, and it does not have some of the new features that we’ll be talking about, like infographics, etc.

There’s the paid version, which is called Pro. This is available to people who pay for Google AI. The AI Pro subscriptions are bundled into that $20 a month subscription for individuals. It is also bundled into Google Workspace AI for companies. For example, Trust Insights is a Google Workspace shop, so everyone who works for us who has a Google account gets NotebookLM the paid version, the $20 a month version.

There’s a third version called NotebookLM Enterprise. This is built for people who are in Google Workspace Enterprise subscriptions. What it does is it’s essentially the exact same software, except that the governance is entirely within your Google Cloud environment. So if you have to meet compliance things like HIPAA and Sock2 and all that stuff, if your Google Workspace environment, the enterprise version, is compliant, then so is your NotebookLM.

So those are the three versions. The paid versions all have the same functionality, so there’s no difference from one to the other—it’s more compliance and paperwork on the enterprise version. This tool is a restrictive compliance, restrictive technology that sits on top of Google’s Gemini model, and it uses Google Gemini 3. Gemini 3 is powering everything that you see in it.

So with that, let’s talk about when you would even use this thing. So Katie, let me ask you this: When do you use this tool?

Katie Robbert – 03:47

I’ve been using it a bunch, and this was actually a question that came up in our Slack community: “What are some of the use cases?” The way that I’ve been using it is to analyze and understand large blocks of qualitative data—customer feedback, voice of customer survey results, things that aren’t necessarily easy to put together a bar chart in a spreadsheet, for example.

We ran a survey, a feedback survey for a client recently about AI usage and what people were doing, and there were a lot of open-ended questions. So to take all of that unstructured text, give it to NotebookLM, and pull out, “Here’s what people are saying, here’s the highlights, here’s what you need to know based on counts,” and it actually did a bit of the analysis, was really helpful. Obviously, as the human, I went back, double checked, and made sure it was correct.

Another use case that I’m using is essentially what you would probably call voice of customer, even though it’s internal to an organization—the Humane Society that I volunteer for. They’re trying to understand what’s working and what’s not working with certain processes and logistics. So we conducted a series of interviews with both staff and volunteers, recorded those interviews, and the transcripts are becoming the source material for NotebookLM.

So now I can query, “What are the top three logistical challenges? What does everyone say is working, or specific to dog care, where are things breaking down?” What was interesting—and I won’t go into the specifics—but what the team thought were the issues were not the issues. As we start to bubble up that information, it’s like, “Oh, these other things over here are actually the issues.”

The reason I like it is because it’s making, at least for me, qualitative analysis so much more straightforward. I like that this is restricted to just the sources versus giving it to something like a Gemini or ChatGPT because I feel like there’s too much extra noise that can go into those systems. Whereas with Notebook, it’s like, “Nope, this is the source.”

Then you know—I know you’ll go through what it all looks like—but you can pick out specific tidbits of information and just do something with that.

Christopher Penn – 06:16

Exactly right. So it is a system that locks into the data that you’re using. So the first and most important thing is it has fewer hallucinations, not none, but has fewer hallucinations because it can draw from the data that you give it.

If you are working on anything that is even marginally risky, for example, finance, law, health, you should probably be getting your own data and putting it into NotebookLM rather than leaving it up to a regular large language model, which, because it’s programmed to be helpful, will be, “I don’t know if it’s right or not, but I’m going to be helpful and I’m going to give you the answer to this question,” which is always delightful.

The second thing is that because you are providing the data, you can—if you don’t have data, it now has the ability to do that. But so, let’s look at a couple of use cases just as a starting point that are really useful. One of them, and it’s this time of year, right? Agencies are at a point where, “Oh yeah, we’ve got to figure out what we did for our clients.” People who are doing performance reviews, “Figure out I’ve got to figure out what I did this year.”

If you have things like, for example, your weekly stand-up meetings or your conference calls, you can put all of those transcripts from your Google Meet, your Zoom, your Firefly Otter, you know, the gazillion and a half companies that all provide these things and ask, “What did we do this year? What are the things that happened? Help me write a performance review.” Because you put the data in, it can generate those things.

So let’s look at a couple of the newer features. First, you can specifically get data either from the web or from Google Drive. So if you wanted to go from the web, one of the things you can do is you can paste in up to 100 links in here, and it will go and fetch those pages for you, which is super handy if you had a list of links that you wanted to grab. It can do that. Now it used to be one at a time.

Katie Robbert – 08:26

I’m trying to think of a use case of when you would do that. I guess I’m thinking back to when we worked at the PR agency and were maybe doing competitive research, for example, and, “Here’s the list of companies that I want you to find out everything about.” Does it grab the specific page or does it grab everything from a site if you give it the URL?

Christopher Penn – 08:51

It grabs the individual URL. So, to that question, you might have an SEO tool like Semrush or Ahrefs or something like that would give you the top pages on your site or competitor site as a list of URLs. Copy-paste, and now you’ve got all that content in here. Super useful. If you wanted to do, like, “What is our competitor’s core messaging?” or “What are the top pages on a competitor’s website? What do they say?” You could bring all that in and then ask the machine for that.

So that’s one set of sources. Another set of sources, which is a huge one for us: Google Drive. We are a Google shop, so when we have documents, for example, like our sales playbook or our many ideal customer profiles, we can now just bring those sources right in and not have to do the whole, print as a PDF and things like that. Super useful.

Third source: YouTube. You can give it YouTube videos and say, “Bring this YouTube video,” and it will transcribe it and make that data available. And of course, you can drag and drop PDFs, text files, Markdown, audio, Word docs, etc. Microsoft Office docs are limited to paid versions, so you can’t do that in the free version.

However, there’s another thing that they added very recently, and that is deep research capabilities. If you have a great deep research prompt, like the Trust Insights Casino framework, you can put that in and you can have it grab the—do the deep research—and it will put in not only the summary deep research report, but it will also grab all of the sources that it used for that report so you can get all the PDFs.

So if you’re saying, “Hey, I want to research, you know, cholesterol for middle-aged Korean men in, for peer-reviewed papers that have DOI numbers,” and you wrote it into a Casino framework, you put that in and it would grab all the papers that it used and put them in. Then, if you wanted, you could just even delete the summary and just work with the papers directly.

Katie Robbert – 10:49

So let me ask you what seems like an obvious question. Why would you do this with Notebook when you can theoretically do this with like Gemini, for example? You can do deep research and then ask it to summarize all the deep research papers and you can talk to the, basically the chat box and say, “What about this? What about this?” I feel like this is very similar. So why would you use NotebookLM versus what everyone’s sort of been accustomed to with Gemini, and Claude, and ChatGPT?

Christopher Penn – 11:23

They don’t get you the sources. They’ll get the URLs to the sources, but they won’t get you the sources. So if your prompt contains something like, “Find me the top 50 papers published on arXiv.org this year about marketing attribution that have been peer reviewed,” and you commission the deep research in here, it will download the PDFs of all those papers for you so that you don’t have to go and do that.

Katie Robbert – 11:46

Got it. I feel like that’s a huge distinction, because then to your point, they become the sources that you pull from, not just information in a report that has links in it.

Christopher Penn – 12:00

Exactly. Because if you read the reports, sometimes depending on the sources pulled, there are hallucinations in it because this receives conflicting information. One of the conditions that creates hallucinations is conflicting information, so that’s one of the things that you have to be very aware of.

By having the sources in here, you can vet them, you can remove the ones that you don’t like, etc., and yank that stuff out. So that’s the input side. It has had a lot of new things that you can do to bring data in, and it’s really, again, those deep research capabilities are super helpful.

A couple of other things have changed. The chat is no longer temporary. Now, for any notebook, if you had a chat, say, “What recipes did the Trust Insights team discuss in 2025 during team meetings?”—I’ll fix the little typo there. If I were to close this window and reopen it, in the old days, it would be a fresh chat.

Now the chat persists so that you can go back and look and say, “Oh, this is what the previous discussion was.” You got sourdough, holiday seasonal recipes, Kelsey is saying she’s making someone’s mac and cheese and so on and so forth. So we obviously know this is accurate; this is representative of our team meetings and things, but this now sticks around.

Katie Robbert – 13:32

So question about history, or maybe this is what you’re going to cover. Let’s say, so we use—we tend to use a shared account. So this is our shared account. Will I see your chat history in the notebook if I use the same notebook?

Christopher Penn – 13:49

If you’re using the same notebook in the same shared account, yes, you will.

Katie Robbert – 13:52

Okay. That’s helpful, especially because one of the use cases that you gave was performance reviews or whatever, something that might be something you don’t necessarily want to share with everybody. Actually, there was a question that came up similar to this of, “I would like to know if I could share a notebook with my students but keep some of the notes to myself, like quiz questions.” I think that’s a really great question and probably a very common use case that people are going to try to use this for.

John Wall – 14:29

Yeah.

Christopher Penn – 14:29

So the answer is you can’t. What’s in the studio, if it’s in the studio, is acceptable to everyone who has access to the notebook if you give them access to the assets.

Skipping ahead slightly, if you go into the share box here, if viewers have access to the chat only, they see none of the assets in the studio. So you could build it so they could have access to just the chat and anything on the right-hand side—any infographics, any videos, any quizzes—they would not see. They only see a chat box. If they see the full notebook, then they see everything, all the assets that have been created. So it’s kind of either-or.

Katie Robbert – 15:05

Then a follow-up question: The chat just scrolls up, correct?

Christopher Penn – 15:09

That’s correct. That’s correct.

Second thing is they now have controls to configure the chat. This is really poorly labeled. There’s Default, there’s Learning Guide, and there’s Custom. Custom gives you 10,000 characters for system instructions. Again, if you’ve done things like the Trust Insights RACE framework and you have the role that you want someone to play, this is where that goes.

You can put in, “You are a James Beard or Michelin-starred chef who is skilled at pastries and pies.” Then, once you put that in here as a role or give it specific instructions—and with 10,000 characters, you can put in things like, “Always respond in Markdown,” or “Always write in active voice.” One of the things people forget is that this is still Gemini under the hood. This is still Gemini.

So all the things you would do in a Gemini chat to condition output, you can put in as system instructions inside NotebookLM. So if I wanted to even take aspects of Virtual Katie, I could put Virtual Katie in here in this notebook—and this is on a per-notebook basis—and have Virtual Katie be in here. Then it can control the response length as well. I typically will default to longer just because I’m usually using this for data extraction, and so I want a longer result. So that’s important. That little setting box is for system instructions.

Katie Robbert – 16:39

Okay. Before we move on from system instructions, it sounds like you’re saying, similar to building like a custom gem, you could sort of do the same thing, but it’s not going to open up and say, “Hey, I’m Virtual Katie, how can I help today?” It’s not exactly that kind of a system.

Christopher Penn – 17:03

That’s right, it’s not a full gem. However, there have been some leaked screenshots from inside Google that is showing that NotebookLM will be available—may be available, I’ll put a big asterisk—may be available inside regular Gemini chats. So if you’re really good at building notebooks and collecting data and cleaning it and getting it ready for analysis, things like that, you may be able to, if those screenshots are true, at some point, connect it to a regular Gemini chat. So it could be a much larger repository than what you can fit inside a gem.

Katie Robbert – 17:42

If I’m following, you can build a proxy custom gem, but it’s not really a custom gem in NotebookLM, but Notebook you need sources for it to work from. Whereas in Gemini, if you build a custom gem, it’s just going to talk to you whether or not the information is correct. Is that sort of an easy distinction?

Christopher Penn – 18:08

Exactly, exactly.

Some other basics in the settings, one thing you can do is you can control the output language. The output language is available in all the languages that Google supports, which is about 86 of them. So, for people who are in the education space, it’s super handy. If you wanted to prepare that notebook for students, you could say, “Hey, if say Polish is your native language,” you can put “Polska” as your language and then it will change the interface and obviously will be available for people in that language.

Now the last part is Studio. Studio used to have four things, then six things, now eight things. What you’re seeing in Studio is a result of the NotebookLM not necessarily being 100% sure what it wants to be when it grows up. It was originally something like a study tool, so mind map, flashcards, reports, and quizzes were the things it shipped with initially, and they were very good.

If I wanted to create a mind map of all the things that we talk about—actually, I’ll pull one up that doesn’t potentially have confidential information issues—and let’s reload this notebook.

Katie Robbert – 19:26

Oh yeah, I mean you can pull up. I think a really good example is like the AI adoption notebook, where you pulled all of the Reddit data, those conversations, right?

Christopher Penn – 19:40

Let’s see if I can go find that.

Katie Robbert – 19:43

That’s in our client services account while you’re pulling that up. Okay.

Christopher Penn – 19:51

So you get things like Mind Maps. Mind maps are one of the most useful ways to traverse a repository, particularly if you put a lot of data in, because you can see what are the big topics within. For each of these topics, once you get to the end node, if you tap on that node, it will then put a query in the chat saying, “What is this node about?” and it will provide a nice machine-generated summary for you.

Katie Robbert – 20:18

I’ve been a big fan of the Mind Maps so far because I feel like it’s a great way to organize the data by topic and subtopic. So, you know, this is problems with AI adoption, digital transformation, the conversations all pulled from Reddit, but that’s a big topic. So it starts to break it down by layers and you can explore each of those layers.

So if I’m using this to create content, for example, I only have to go into one of the layers without having to do all of the querying of the system to try to figure out what those layers of topics are. It’s already told me, “Here’s the path that you follow to get to what it is you should be talking about.”

Christopher Penn – 21:00

Exactly. So those are the basic study ones. Any icon where you have a pencil next to it means you can prompt it. Those prompts can be about 2,000 characters long. So if you’re going to make flashcards, for example, you could choose how many cards and how difficult you want them to be, but then add conditioning instructions for the generation.

Now, the two that we covered in the past that were a lot of fun, Audio Overview and Video Overview, those are essentially things that create miniature podcasts with synthetic hosts and miniature videos. The videos have gotten substantially better since the last show because they now use Google’s Nano Banana Pro—the silliest named system you can possibly imagine. Let’s take a look at… actually, no, I don’t have a video set up on this one for that particular use case.

The two newest ones are Infographic and Slide Deck. This again uses Nano Banana Pro. These are not available in the free version; you will see them grayed out because they are fairly intensive. What they can do is, again, creating any kind of infographic—landscape, portrait, square—in any language with any level of detail.

So here’s a fun one: If you want to just take your LinkedIn profile, put it in as a source document and then create an infographic of your career. This is Katie and her career progression and how and what has happened over the years. Look at this: “Drove significant business as director, grew the team by 50%, increased revenues 54% year over year.” I mean, Katie’s a rock star.

Katie Robbert – 22:42

All in an infographic.

Christopher Penn – 22:44

All in an infographic.

Katie Robbert – 22:45

But you know, it’s sort of a silly example, but if you think about the differentiation in the job market, how do you stand out? How do you show someone? You have like a millisecond to make a first impression. This is a great way to do it.

It’s also a really good way to see if the information in your resume or LinkedIn profile is actually meaningful. If you just have, “Hey, I’m John Wall and I’ve hosted Marketing over Coffee for 8,000 years,” and that’s it—no offense, John, I know that your LinkedIn profile is much deeper than that—but if that’s all you have listed in there and you’re wondering why nobody’s hiring you, this is going to be a really good way to, for lack of a better term, QA your resume and your profile to see if, “Hey, there’s nothing in here of substance. This is why everybody’s passing over.” I know a lot of people are looking for jobs right now, so this is a great way to be sort of gut-checking, “Does my LinkedIn profile or my resume even say anything?”

You know, mine, you could argue that there’s plenty of room for improvement in mine, but I feel like overall it kind of tells you what I do.

Christopher Penn – 24:05

And it definitely has that nice “up and to the right” feeling of, like, “This is someone who’s a rock star.” Our chat agrees, by the way. People are like, “Yes, our chat agrees. Everyone says, ‘Go, Katie.'”

Now, at a slightly more serious tone, the tools can generate content, audios, and videos, as well as infographics in pretty much any application. So this is the extremely boring 834-page Federal Student Aid Handbook published by the U.S. Department of Education that nobody likes to read. I used—I did this at a financial aid conference a couple weeks ago. In this, in the notebook, I said, “Okay, I need to create an explainer for a parent on how to apply for a Pell Grant.” It goes through and it does a really good job.

It explains all the conditions and stuff for what it would take to get through a Pell Grant. Now here’s the really useful thing, and this is the thing that all the financial aid administrators in the room just went gaga over: eighty-six languages. In Massachusetts, because it was a Massachusetts conference, 1.2% of our students come from Haitian Creole families that speak Haitian Creole. Zero percent of financial aid administrators speak Haitian Creole, and certainly not at a level that they would be comfortable in having a conversation with the student’s family.

So what this tool does is it spits out Haitian Creole—that you obviously fact-check if it’s something sensitive. But I had somebody who did. They’re like, “Yeah, this explains exactly what’s in the English video, but in a language that we don’t have access to as individuals,” and it creates really, really good content that explains the process, and this is of getting, “How to apply for a Pell Grant.”

Katie Robbert – 26:00

So we do have a comment, Chris. With a free account, you can create an infographic or a slide deck, but you’re limited to one or two a day, which I think, depending on what you’re doing, is okay because you’re still getting something, and you can be really thoughtful and strategic about how you’re using the system.

John, I want to check in with you. Is NotebookLM something that you’ve been using for Marketing over Coffee transcripts? I could see you building out some sort of end-of-year review. We all just got our Spotify year-in-reviews and our listening ages, which are insulting. But I could see you doing a Marketing over Coffee end-of-year review of, “Here’s what we talked about, here’s how many guests we had, here’s how we,” you know, putting your stats together. Now I’m just adding more to your to-do list.

John Wall – 26:50

Yeah, right. The big thing for Marketing over Coffee is book summaries. I get so many business books and marketing books, and people want either to have them blurbed or get on the show or whatever. It’s just completely fantastic to load the thing up in there, and I don’t—because it’s a private workspace—I don’t have to worry about the book getting out into the rest of the world.

But I can say, “Hey, summarize the five chapters.” Or the big one is, “Tell me what’s in this book that hasn’t already been written about before.” You can see if a book is for real and if it has any juice. So yeah, NotebookLM is just all about boiling down huge information resources into stuff that I can use.

The other one is just, you know, any kind of trying to figure out a problem, whether it’s mechanical or coding or whatever, where you just upload the five user manuals and now you can just ask questions and just learn about, get the answer to your problem. You don’t have to comb through five indexes and continually dig to get to where you want to go.

So I haven’t played around with the transcript stuff. The stuff we’ve been using so far has gotten better and better, but yeah, I think it is time to do a toe-to-toe showdown and see how good it’s doing versus some of the other. In fact, because I do still have a couple of tools on the payroll like that, it may be time to jettison some of that stuff if it’s doing a better job and can handle some of the context stuff that the existing tools can’t.

Katie Robbert – 28:20

I think you bring up a really good point. With a lot of these tools getting more and more sophisticated, I would say probably at least quarterly for a lot of companies, but you know, every six months or so, you should be evaluating the things that you’re paying for and can some of these tools that come along with your Google Workspace account, or whatever, replace a lot of the vendors? The vendors aren’t going to be happy, but that’s just more motivation for them to get it together.

Christopher Penn – 28:54

One of the biggest questions people have is, “How safe is my data?” Google actually has its own privacy policy just for NotebookLM. They say, “We do not train on the data unless you provide feedback.” So, if you press the thumbs up, thumbs down in the system, then you’re submitting that for training. Otherwise, for the consumer version, your data is not used to train models.

For the paid versions, if they’re Workspace, if they’re within Google Workspace, that is governed by your overall Google Workspace terms of service. Education and Enterprise have an even stricter policy. So those ones, like there’s not even human review in the highest levels of security, but across the board, including for the free version, Google says, “We do not train on the data you put in NotebookLM.”

It looks like we have a bunch of questions here. So Brittany’s asking, “Is there a way to run a year-in-review analysis on a news site without having to paste every individual article URL?”

Use the deep research facility for that and say, “Grab these articles.” You can also, if you have a Semrush, Ahrefs, Moz account that can look at that news site, you can say, “Give me the URLs, the top 50 or 100 URLs based on page traffic for that,” and bring that in.

Katie Robbert – 30:16

The next question—we were just talking about the privacy—is, “Wouldn’t you agree that the paid version should be used for private information, even proprietary books, meaning info we don’t want the models to be trained on, or is that not accurate?”

Christopher Penn – 30:32

Yeah, so as it says on screen for NotebookLM specifically, which is different than Google’s regular privacy policy, they do not train on NotebookLM data, which makes sense because otherwise no one would use it.

Katie Robbert – 30:45

Another question that came up—we were talking about this a little bit earlier, but I think it’s good to further clarify—is, “I would love to understand clear use case options of building a gem versus NotebookLM. When should I use one versus the other? Advantages, disadvantages.”

I think what we’re saying is one of the big advantages of NotebookLM is that you have to have all of the sources. So we talk a lot about the better generative AI: Generative AI works better with more data and context. So the more knowledge blocks, the more deep research, the more context you can give it, the better it’s going to work. Chris, you have something highlighted.

Christopher Penn – 31:28

Yeah, no, exactly. What you’re saying is NotebookLM is designed to answer questions based on the information you provided. Gems will use knowledge in the knowledge base as part of the prompt, but they are not locked to that information, which means that you can get those hallucinations. Or they can also—and this happens frequently with Gemini—it will trigger its own web search. Well, for certain topics like finance, law, and health, I don’t want it going to search the web. I don’t want it pulling information from Esther’s Healing Crystal blog. I don’t want that. I want guaranteed good information.

For example, I was doing some research recently for myself on cholesterol. I wanted to pull in only peer-reviewed papers from the last five years, specifically from the Korean Ministry of Health because I’m Korean genetically by birth. So I wanted to say, “What did that country’s health authorities have to say about this, about having high cholesterol?” So I grabbed just that data and I wanted Google to answer questions only from that data. When you use regular Gemini, even in a gem, it was still getting cross-contaminated with U.S. American standard health standards, which are not the same.

Katie Robbert – 32:50

One of the use cases that we shared with a client of ours: They were looking at different contracts. So what’s different? I think a really good use case is basically if you want to sort of call it spot the difference. You can give it your 2024 contract and your 2025 contract into NotebookLM and say, “Tell me what’s different about these two contracts.” That’s again a really good use case because Notebook is only going to use the sources; it’s not going to hallucinate and say, “Well, I found this thing way over here.”

Great example of hallucination: I was working on our workshop landing page the other day, and it started giving me formulas for long division, and I was just very confused. I’m like, “Why are we talking about this finding the square root of eight when I’m asking you to put my workshop page into two columns in HTML?” It was off the rails. You’re going to get less of that—not completely zero, but less of that—in a NotebookLM.

Christopher Penn – 33:58

Yep, another use case very similar to that. I was asked recently by you, Katie, actually, “Hey Chris, you’ve got a lot of content out there under your own channel. We’d love to have—it makes sense to take stuff that our ideal customers would find useful and put it on Trust Insights channels.”

So I pulled my entire YouTube history for two years. There’s a free, pain-in-the-butt utility called ytdlp that is a command-line tool. If you Google for it, it’s called ytdlp; it is on GitHub. So you’ve got to be super comfortable with the terminal because that’s what it is, but it can download all the closed captions for any YouTube video.

So I put that in and then put all 700 episodes in because it contains my newsletter and my daily videos. I put in our sales playbook, which we covered on a previous episode of the live stream. Our sales playbook has five different ICPs, and I said, “Of the content that I’ve made on my personal channel, which content is the best fit for our ideal customer profiles? Which profile, which content, and why?” It was able to come up with a nice long list, and now we have a punch list of, “Okay, we’re going to make Trust Insights versions of these,” because also a lot of my personal stuff is fairly unpolished. Like, most of the videos now are me cooking in my kitchen while I’m answering questions so I can multitask on a Sunday morning. That’s not the level of professional…

Katie Robbert – 35:25

That is not the Trust Insights brand standards that we strive for.

Christopher Penn – 35:30

No, it’s me flinging stuff out of the kitchen, but it does answer the questions. However, we’re going to actually cover this on a future episode of the live stream when we talk about Google Workspace Studio and how to automate that process, but that’s for another time. But this is a use case of NotebookLM where it can take what is very close to—I think I counted—three million words. Three million words. It exceeds the context window of any AI model and is able to process it very well.

Katie Robbert – 36:04

So another question that came up, one more if you had a chance: “Do you find Gemini NotebookLM’s deep research gets blocked by online sources any more or less than ChatGPT?”

Christopher Penn – 36:15

Less. The reason why is because Gemini is Google, and Google has Googlebot, and so many sites have made exceptions to their crawling policies for Google so that Google can index them. Google has basically conditioned content owners for the last 30 years to say, “Let us have your stuff.”

Now Gemini piggybacks off of the Google infrastructure and can see things that a lot of companies have just outright blocked anything from OpenAI, like, “Oh, ChatGPT, we don’t want that here,” not realizing that Google literally has all the information. This is one of the reasons why Google is such a huge threat to OpenAI because they have 30 years of everything. They have your Gmail, they have your YouTube, they have your Google Docs, they have Search, they have it all. Google Scholar.

They have greater access to data, and they have cached versions of that. When you go into a Google search result in the old-fashioned, old days, you could always tap on, “Show me the cached version of this page that Google previously saw.” Guess where that lives? In Google’s data center. And guess what? Gemini can see anything in Google’s data centers. It’s theirs.

Katie Robbert – 37:35

Gotcha. That’s incredibly helpful.

A question that came up in our Slack community, or more of a feature request—Google, if you’re listening—is, “It would be great to have an export feature just for Notes.” We haven’t even talked about that feature of what it looks like.

So let’s take a step back before we get into how to export support it. Chris, in this example, you’ve uploaded or you’ve added sources from all of your YouTube videos—those are all the transcripts. You’ve also added the sales playbook, which are the ICPs. Then you asked basically the question of, “Which of the videos that I’ve created are the most aligned with the Trust Insights ICP?”

It’s giving you a response. Something that you can do in NotebookLM is save that to a Note. So in the bottom, you have next to the copy, the up, the down, the thumbs up, thumbs down, you can save it as a note. This is the feature that I really like because you can basically—you’re basically saving the responses to each individual query.

Then those individual responses, those notes, can then be converted to a Source. So you can use just that information so it becomes very nested within itself. You can convert that to a source and then create a report just on that source or an infographic just on that source, bypassing the, you know, three million words that Chris has already added. You can look at just that thing, which I think is really helpful.

Back to one of my use cases that I did for one of our clients in analyzing their feedback survey, I went through question by question. The responses to each of those questions became a note and then became a source. Then I could go through with the client’s branding guidelines, taking the branding guidelines and the source, and create an infographic just for that set of responses, turning it basically into an internally or externally polished slide deck of more interesting data than just your standard bar charts.

Christopher Penn – 39:57

Yep. Now, here’s the catch: It does not have otherwise robust stuff. When you hit this copy button, it basically takes the styled Markdown here and puts it into your clipboard. That’s it.

Katie Robbert – 40:12

Yeah, well, and so that goes back to the feature request of, “It would be nice if you could export an individual note.”

Christopher Penn – 40:21

Exactly. So hopefully that’s on the way. If not, if it does in fact end up getting integrated into regular Gemini, it will be something that then Gemini could put into its canvas and restructure.

Katie Robbert – 40:44

So I just created a graphic of my career like Chris mentioned, and it came out great. I will post on LinkedIn. Feel free to tag us. We’d love to see it.

Christopher Penn – 40:54

One thing that is very useful when you’re doing infographics, remember this is Gemini. Gemini, if you provide information about what a good infographic is, it will do better. So for example, I very recently was working on this because I was like, “I don’t love some of the infographics that it comes up with.” So I commissioned a deep research report called the Architectonics of Information. This is just your standard deep research report from regular Gemini on what constitutes good infographics these days.

So now if I load this as a source, I can then refer to it in the infographic construction and have it be styled along this way. Same with brand style guidelines. You can load brand style guidelines into it for infographics and slide decks and things, and it will reasonably comply.

Not always and not reliably, but reasonably, it does a reasonably good job, particularly for things like colors and stuff, because under the hood it’s still Nano Banana Pro.

Katie Robbert – 42:01

And so, as I mentioned, when I did the feedback survey for the client, we used their brand guidelines for each of the infographics and it came out fairly decent enough that the client knew that it was their branding and they were happy with the results.

Sort of a side note, as you said, “So I commissioned a deep research report.” I feel like you’re very much the Hermione of the group, where she was like, “Well, I was curious and I went to the library,” and everybody knows that’s going to be her first stop. So for your first stop is a deep research report, which I don’t mean as an insult. I think it’s a great best practice because we don’t know everything about everything.

If you’re not happy with the results you’re getting, there are ways to fix it. This actually came up in our Slack community. Again, if you want to join the conversation over in our free Slack community, you can do so at TrustInsights.ai/analyticsformarketers. We’ve been discussing this NotebookLM and the new features today. Somebody said, “I don’t love some of the infographics, how do I fix that?” One is you edit, you prompt it, and then two, you can give it those guidelines for what a good infographic should be.

Christopher Penn – 43:21

In fact, we’re going to go ahead and put in that PDF because it was just a deep research. We’ll put that PDF in Analytics for Marketers. If you’d like a copy of it, you can get it in the Slack group. We’ll have it just after the show ends.

The thing about NotebookLM, just kind of to put a bow on stuff, is it really is a combination of a library that you build combined with a data processor to crank out stuff. There are some things it can’t do. It cannot do any form of math. It’s still a language model and it does not have access to Gemini’s coding facilities.

Regular Gemini now writes code in the background. It doesn’t tell you that it is, but it is doing that. So even in regular Gemini, about half of the time it can write and execute code. It’s unreliable. Google Colab, which we’ve covered on previous episodes, writes code 100% of the time, and that code typically executes properly. So it does math well.

You cannot do math in NotebookLM. So if you are using stuff that you want to have for mathematical conclusions, you need to provide the finished data in. In the example of Katie’s career infographic, those numbers were not computed; those were numbers straight off of her LinkedIn profile. So it cannot do math. So don’t try to have it do math. It ends very badly.

Otherwise, it is Gemini with restrictions, with good restrictions. So treat it as such. The same things that you do in regular Gemini—the RACE framework, “Ask me one question at a time until you have enough information to successfully complete the task,” “Recap this,” as system instructions and things—it understands how to do, particularly for the output products. When you say infographic, don’t go, “Make me an infographic,” right here. That’s going to make garbage.

It can handle thousands of characters, so take the time to think through what a good prompt would be for this. Or even better, in regular Gemini, build a gem with some deep research that says, “Help me write prompts to generate great infographics.” Then you can just go to your gem and say, “Here’s what I want to do,” and have it improve those things because it is still just Gemini.

Katie Robbert – 45:36

John, what are you going to do first?

John Wall – 45:39

What am I going to do first? We didn’t even get into there’s a bunch of these agents in the Workspace Studio too, which is a little off-track, but that’s what’s got me interested right now because there’s a bunch of just automation for your inbox and a bunch of other things over there. So that’s number one on the list.

Katie Robbert – 45:58

Well, I think you should stay tuned slash attend our upcoming live stream where we cover that probably in a couple of weeks.

John Wall – 46:07

Sounds like a plan.

Katie Robbert – 46:10

One final question Nancy was asking, “Was NotebookLM able to access Katie’s LinkedIn URL?” No, I exported the resume, the PDF, the LinkedIn profile and gave that as a source to NotebookLM.

Christopher Penn – 46:25

There is a secondary hack for that. You can’t take the…

John Wall – 46:31

You can’t.

Christopher Penn – 46:31

NotebookLM does not handle any form of video, but regular Gemini does. So one of my favorite hacks is to open up my phone, turn on screen recording, go to LinkedIn, browse what I want to browse. Maybe I’m searching on a hashtag, maybe I’m just looking at what Katie’s posts are and things like that. Then browse, then turn off screen recording, put the video into regular Gemini and have it create a transcript of everything it sees. That will give you the data you need that you can then put into NotebookLM.

Katie Robbert – 47:00

We have covered that in previous episodes. So we keep showing this URL, TrustInsights.ai/youtube, specifically the So What playlist. That’s all of our past live streams. That’s where you’ll find a lot of this information because we try to be really thorough and cover as much of this as humanly possible. If there’s something that you still have questions about that we haven’t covered, you can join our free Slack community Analytics for Marketers at TrustInsights.ai/analyticsformarketers. It’s free to join. We’re there every day except for weekends, but even sometimes Chris is there as well.

Christopher Penn – 47:34

All right folks, that’s going to do it for this week’s So What. Thanks for tuning in, and we will catch you all on the next one. Thanks for watching today. Be sure to subscribe to our show wherever you’re watching it. For more resources and to learn more, check out the Trust Insights podcast at TrustInsights.ai/tipodcast and a weekly email newsletter at TrustInsights.ai/newsletter. Got questions about what you saw in today’s episode? Join our free Analytics for Marketers Slack Group at TrustInsights.ai/analyticsformarketers. See you next time.

Need help with your marketing AI and analytics?

So What? Marketing Analytics and Insights Live

airs every Thursday at 1 pm EST.

In this episode you’ll learn:

Transcript:

Leave a Reply Cancel reply

Pin It on Pinterest