Two podcast hosts banter back and forth during the final episode of their series, audibly eager to share heartbreaking stories with listeners. “We were informed by the producers of the show that we are not human,” a male voice stutters, the problem in the middle. The conversation between the bot and its female voice counterpart turns awkward after that—an interesting, if misleading, example of Google’s NotebookLM tool, and its AI testing podcasts.
Audio of the conversation was shared on Reddit over the weekend. The original poster admits in the comments section that they fed the NotebookLM software instructions for the AI voices to play the role of this pseudo-disruption. So, there is no feeling; AI bots don’t know themselves yet. However, many users in tech magazines, on TikTok, and elsewhere are recommending convincing AI podcasts, which are generated from uploaded documents with an Audio Review feature.
“The magic of this tool is that people get to listen to something they can’t get on YouTube or an existing podcast,” said Raiza Martin, who leads the NotebookLM team within Google Labs. Martin talks about uploading a 100-slide deck of commercials to the tool and listening to an eight-minute podcast summary as he multitasks.
First introduced last year, NotebookLM is an online research assistant with features common to AI software tools, such as document summarization. But it’s the audio overview option, released in September, that’s taking the internet by storm. Online users are sharing snippets of their AI-generated podcasts made from Goldman Sachs data dumps and testing the tools’ limits with stunts, like just repeatedly uploading the words “poop” and “fart.” Still confused? Here’s what you need to know.
Generating That AI Podcast
Audio overview is a fun AI feature to try out, because it doesn’t cost the user anything—all you need is a Google login. Start by logging into your personal account and visiting the NotebookLM website. Click the readable plus arrow A new Notebook to start loading your source material.
Each Notebook can work with up to 50 source documents, and these do not need to be stored on your computer. Google Docs and Slides are easy to import. You can also upload websites and YouTube videos, keeping in mind some caveats. Only text from websites will be analyzed, not images or layout, and content cannot be paid. For YouTube, the Notebook will automatically use the transcript and the linked videos must be public.
After you’ve extracted all your links and documents, you’ll want to open the file A notebook located in the lower right corner of the screen. get the Audio Overview and then click on Produce button. Next, you’ll need to be patient, because it can take a few minutes to load, depending on how much source you’re using.
After the tool generates an AI podcast, you can create a shardable link to the audio or simply download the file. Additionally, you have the option to adjust its playback speed, in case you need the podcast to be faster or slowed down more.
The Future of AI Podcasts
The Internet has gotten creative with NotebookLM’s audio feature, using it to create audio-based “dive-in-depth” audio-based dives into complex technical topics, produce concise summaries of dense research papers, and produce “podcasts” about personal health and fitness routines. Which raises an important question: You should using NotebookLM to check your personal files more?
The snapshots generated on NotebookLM, according to Google spokesperson Justin Burr, “are based entirely on the source uploaded by the user. That is, your personal data is not used to train NotebookLM, so any private or sensitive information you have on your sources will remain private , unless you choose to share your sources with collaborators.” This seems to be one of the benefits of Google labeling NotebookLM; hearing Google’s framework for it, the company is currently gathering feedback on the product, it’s fast and responsive, and NotebookLM is out of its multibillion-dollar ad business ! In the meantime.