During the final episode of their series, two podcast hosts engaged in a conversation marked by evident anxiety as they prepared to share distressing news with their listeners. “We were, uh, informed by the show’s producers that we’re not human,” a male-sounding voice revealed, clearly in the throes of an existential crisis. The discussion between the bot and his female-sounding cohost subsequently became more uncomfortable, illustrating an engaging yet deceptive example of Google’s NotebookLM tool and its experimental AI podcasts.
Over the weekend, audio of this conversation gained viral traction on Reddit. The original poster disclosed in the comments section that instructions were fed to the NotebookLM software for the AI voices to enact this pseudo-freakout. Contrary to causing alarm, this confirmed that the AI bots had not attained sentience. Instead, the convincing nature of these AI-generated podcasts received praise from users in tech media, on TikTok, and other platforms, demonstrating the capabilities of the Audio Overviews feature to create engaging content from uploaded documents.
Raiza Martin, who heads the NotebookLM team within Google Labs, emphasized the unique experience provided by this tool, remarking, “The magic of the tool is that people get to listen to something that they ordinarily would not be able to just find on YouTube or an existing podcast.” Martin shared her recent experience of converting a 100-slide commercialization deck into an 8-minute podcast summary, which she found useful while multitasking.
NotebookLM, introduced last year, is an online research assistant featuring standard AI functionalities, such as document summarization. However, it is the Audio Overviews option, released in September, that has caught the public’s attention. Online users have shared snippets of AI-generated podcasts created from various sources, including data sets from Goldman Sachs, and engaged in playful testing by uploading repetitive words.
To generate an AI podcast using Audio Overviews, users need only a Google login. By signing into their personal account and accessing the NotebookLM website, users can start a new notebook and upload up to 50 source documents, including files from Google Docs and Slides, websites, and public YouTube video transcripts. The tool analyzes text content without considering images or layouts, and can only access non-paywalled stories.
Once the links and documents are uploaded, users can navigate to the Audio Overview section and click Generate. The process may take a few minutes, depending on the volume of source material. After generation, users can share the audio via a link or download the file, and adjust playback speed as needed.
NotebookLM’s audio feature has enabled internet users to get creative, producing deep dives into complex topics, summarizing dense research papers, and creating “podcasts” about personal interests. This raises the issue of whether the tool should be used for personal files. According to Google spokesperson Justin Burr, the summaries generated are based solely on the uploaded source material, ensuring that personal data remains private unless shared with collaborators. Burr noted that this characteristic is part of Google’s experimental approach to NotebookLM, highlighting the company’s current emphasis on gathering feedback and refining the product independent of its larger ad business operations.
