The Real Value of Google NotebookLM
The instant podcasts got us excited, but the real genius of Google's buzzy tool is how easily it applies the power of AI to data.
Three months ago, Google opened our minds to something new: applying the power of generative AI to create instant podcasts about any topic we wanted. All we had to supply was the raw material.
NotebookLM is definitely having a moment, especially among journalists and communications professionals. The audio overviews feature put the tool on the map, and the voices of its chatty synthetic hosts are now almost as recognizable as Siri and Alexa. However, NotebookLM's ultimate impact may have less to do with those audio overviews than you think.
Before I get to that, don't forget that the early-bird deadline for my 2025 course, AI for PR & Media Professionals, is approaching fast. Be sure to register before December 15 with the code EARLYBIRD to get a 25% discount. The prices go up the next day, so you'd better grab your spot as soon as possible. Want to learn more about the class? Immerse yourself here.
A MESSAGE FROM ADOBE
Here’s something you might not know: New features in Adobe Stock enable creators and media folks like you to accelerate your content selection process and find more unique, relevant content you can customize to your aesthetic and brand preferences.
Powered by Adobe Firefly, you can now customize stock images before purchasing:
Swap or remove backgrounds with a click.
Expand images to fit any format.
Generate variations that match your chosen image's style or composition.
Just type a prompt, and watch as AI transforms your stock picks (while maintaining their professional quality).
The coolest part? Original contributors earn when their images inspire AI-generated content. Its professional stock imagery meets generative AI, designed to speed up your creative workflow. Check it out here.
The NotebookLM Effect
Not many AI products have caught on quite as quickly as Google NotebookLM. Which is strange, because the feature had actually been around for months before it found fame as an automatic podcast generator. NotebookLM was originally an easy way to apply the power of Gemini to a set of files.
At first, NotebookLM existed as an interesting but unremarkable product among the myriad features, models, and hyped-up products in the Google Gemini Cinematic Universe. You could feed it some files and ask Gemini about them, sure, but you could also do that in a basic Gemini chat. The idea didn't seem that novel — that is, until Google introduced audio overviews, allowing users to instantly create a podcast about those files, describing the content within them in detail via a couple of smart-sounding synthetic voices.
The audio deep dives were a sensation. Just when we thought we were over the pure "wonderment" of AI, Google's synthetic conversations gave us permission to let our minds be blown again. NotebookLM's audio overviews were the talk of the town at industry conferences, social media threads, and, well, podcasts, when they debuted in September.
You can see why. If ChatGPT showed people that AI could author content, NotebookLM made them aware of its ability to convert it to something else with ease: Hit a button, get your article as a podcast. Once the feature caught fire, Google quickly shipped an upgrade that let users steer the focus of the conversation. (Several other enhancements are apparently on Google's road map, according to former NotebookLM Product Manager Raiza Martin and Editorial Director Steven Johnson, speaking on the Google DeepMind podcast.)
The big irony with all this hype over audio overviews is that, of all the features of NotebookLM, it's probably the one with the least utility.
Don't believe me? Check out the AI Horizon podcast, one of what I presume is an entirely new class of podcasts that are essentially a series of Google NotebookLM conversations in an RSS feed. To be clear, this is a real, downloadable podcast that you can find on Apple, Spotify, and everywhere else.
How to Get the Most Out of Google NotebookLM
You only need to listen to two or three episodes to begin to spot the flaws in NotebookLM audio overviews. While Google has expertly tailored the emotion, inflections, and vocabulary of the two hosts, it hasn't quite gotten the rhythm right — especially if they're talking for more than 10 minutes. It's all chit-chat, all the time, which manifests as them passing the conversational baton to each other too often. Since neither one can ever get a meaty thought in, it keeps everything decidedly surface level.
The longer you listen, the more their habits start to irritate you. After a while, the banter gets unnatural, too smooth. They finish each other's sentences in ways humans never would. They say "exactly" and supportive phrases too often. On top of that, the summarization engine is careless: This episode about "AI in 13 Charts" repeats whole sections of what it talked about earlier in the conversation, just slightly tweaked (compare the discussion at 7:35 with 13:15).
The True Power of NotebookLM
To be fair, Google never said its audio overviews were ever going to become the next Serial. They're meant to simply give a summary and some guidance on the data in the folder. So if you put aside the compelling idea of reformatting content, you start to realize the real power of NotebookLM is how it applies generative abilities to custom data sets.
Using AI with specific data is a huge area of focus for both companies and consumers. Much of the discussion in enterprise AI is about how to leverage it with an organization's proprietary data. For individuals, a Holy Grail is for AI to act as a true "digital twin," and for that it would need access to all manner of personal data. Doing either of these concepts safely, privately, and at scale is complicated.
Google NotebookLM simplifies the idea by scaling back on the ambition. Put aside the entirety of the company's data; just do one project. Same with the personal: Apply the power of AI to a single area of your life, and not only are you more likely to get a good result, you'll also be more comfortable doing it.
Applying AI to data is hardly novel — you can do it within any chatbot simply by attaching some files. The genius of NotebookLM is reversing the order of that interaction: By putting the files first and that chat second, it invites returning often to the same data over and over. This is where NotebookLM does better than Claude Projects, which is essentially the same idea, but less intuitive. Even something as simple as being able to click the boxes of the documents you want to reference in a specific interaction is important, adding a level of precision that's hard to replicate in an ongoing chat in the regular Gemini chatbot.
Journalists on a beat can find a lot of utility in the smart-notebook idea. If you're covering an ongoing court case (say, an antitrust action against a major tech company), you could upload all the court documents, your notes, and past coverage. Then you can converse with some or all of those things to discover insights in minutes that would have previously taken hours or days.
Again, this idea isn't new, but NotebookLM takes it to a new level of convenience. The marriage of a simple folder system with AI is easy to understand, and since it's powered by natural language, there are infinite ways to hone your use of it. This list of tips for journalists from Chris Moran, the Guardian's head of newsroom innovation is a good place to start.
NotebookLM is far from perfect, however. It's not as collaborative as it could be — questions you ask and answers you get aren't visible to others with access. Google says it doesn't train on your data in NotebookLM, but those concerned about privacy may want to consider an enterprise alternative like Elvex or a custom solution built in a tool like MindStudio, at least until NotebookLM has its own enterprise version (said to be in the works).
That may take a little longer than expected. Three key people on the team, including product manager Raiza Martin, recently left Google to start their own company, which will surely slow down development and allow competitors, such as AnyTopic, to catch up. Nonetheless, NotebookLM will go down in history not just as the AI product that simplified using AI with data, but also as the one that finally got us using Gemini.