Google's NotebookLM now allows you to customize your AI podcasts.

Google has introduced a new personalization tool for the popular artificial intelligence podcasts in its NotebookLM software. I had early access and decided to test it using "The Metamorphosis" by Franz Kafka as source material, dedicating several hours to generate podcasts about this foundational work, some of which turned out to be quite peculiar.

Launched by Google Labs in 2023 as an experimental writing tool focused on artificial intelligence, NotebookLM has seen a resurgence in user interest since early September, when developers added the option to generate podcast-style conversations between two AI voices—one male and the other female—based on uploaded documents. Although these "deep analyses" in audio can be used for study and productivity, many of the viral clips online focused on the entertainment factor of having robot hosts discuss strange or highly personal documents, like a LinkedIn profile.

Raiza Martin, the lead of the NotebookLM team at Google Labs, expresses enthusiasm for offering users more control over the content of these synthetic podcasts. “It's the most requested feature we've heard,” she says. “They want to provide some feedback on what the analysis should focus on.” According to Martin, this update is just the first of many changes to come.

With the first anniversary of its full launch in sight, NotebookLM has also removed the "experimental" label, indicating that it isn’t heading toward the well-known fate of abandoned Google software, at least for the moment. Martin notes that the removal of this label is due to the team reaching internal milestones regarding overall quality, user retention, and interface standards. Additionally, she mentions that users can expect a greater level of stability in the software.

To create an AI podcast using NotebookLM, one should open the Google Labs website and start a new notebook. Then, documents that you want to use for the audio output must be added, which can be files from the computer or YouTube links. By clicking on the notebook guide, the option to generate a deep analysis and the option to customize first will now be visible. You can choose "Customize" and add the request for how you would like the AI podcast to turn out. The software suggests considering which sections of the sources to highlight, which broader themes to explore, or different audiences to target with the message.

One tip Martin shares for trying the new feature is to generate the “Audio Summary” unchanged and, while listening to this first version, note any questions or topics you would like to see expanded. Afterwards, you can use these notes as a starting point to create requests in NotebookLM and regenerate that AI podcast with personal interests in mind.

By uploading an 80-page file of Kafka’s famous work, I saw how personalization would work for NotebookLM users. The first Audio Summary it generated, without any customization in the request, was a fairly solid yet broad overview of what happens in the novel, along with some discussions of its key themes. It wasn’t groundbreaking, but it was decent.

Thinking like a literature student, my first adjustment in the request was to ask for the podcast discussion to focus more on the themes of alienation and oppressive bureaucracy found in the book. With this additional tweak, NotebookLM did a good job addressing these motifs, generating a discussion similar to what one might hear in university classrooms. While it was somewhat meandering, it was entirely pleasant to listen to.

Then, I asked the tool to focus on specific pages of the source document. This result was more disappointing than the previous tests, as it sounded quite similar to the podcast generated without a personalized request. However, if I had uploaded more material to NotebookLM, perhaps the instructions to focus on specific aspects would have produced more satisfying results.

The most interesting way to adjust the AI podcasts, in my experience, was to ask the synthetic hosts to address diverse audiences. By requesting that they explain the novel to a recent graduate starting their first office job, the hosts insightfully discussed what it means to go through major life changes. Nonetheless, I nearly spilled my coffee when I heard NotebookLM’s attempt to explain Kafka's prose to a lively group of drag queens. “Darling, let's talk about loneliness,” the male AI voice says. “Because Kafka doesn’t hold back.”

Although some of the expressions used by the hosts when communicating with the drag queens were somewhat confusing and a bit awkward, I was surprised to see how the content was adapted to highlight aspects of the novel that queer individuals might identify with, such as Kafka’s approach to otherness and familial tensions. While NotebookLM may simplify the details of an extensive document or confuse some of them, the ability to generate more personalized podcasts from diverse sources really seems to represent a transformation—and fortunately, nothing like turning into a giant insect.