Google on Thursday updated the audio summarization feature of its AI-powered research and note-taking assistant NotebookLM, which recently attracted a lot of attention for its podcast-like audio conversations based on content users share, with the ability to guide those conversations and focus on specific topics rather than simply generating holistic audio summaries.
Today, audio overviews in NotebookLM allow users to digest and understand information in long documents or videos through AI-generated audio conversations. Shortly after its launch last month, the feature helped NotebookLM gain attention as many began sharing audio summaries of their content on social media, including those created with their journals or diaries.
While Google has not revealed the traction NotebookLM has received as a result, data from website traffic analytics platform SimilarWeb suggests that NotebookLM saw a more than 371% increase in its traffic in September to 3.07 million. monthly visits, compared to 652,181 a month ago.
Until now, Audio Overviews automatically generated AI conversations from user feeds. But since conversations sometimes revolve around content that isn't important, Google is introducing an update that lets you customize overviews to your needs. This allows users to make the audio focus more on an explicit topic within their content.
A dedicated “Custom” control is available before the existing “Generate” button to allow you to provide instructions for AI hosts in the audio to focus on a specific point.
Raiza Martin, NotebookLM product lead and senior AI product manager at Google Labs, told TechCrunch that the update gives users a way to nudge AI to move in the direction they want.
“The entire team has dedicated itself to listening and analyzing all the comments we have received. And the number one feature that came out that people wanted was just giving AI a little push,” he said.
Personalizing audio summaries can also help reduce hallucinations to a certain extent, that is, those occasions when the AI creates the content on its own. However, Martin said the NotebookLM team tracks user feedback and tries to detect hallucinations as quickly as possible.
He also emphasized that customizing audio summaries does not mean that user instructions will be used to train the AI model.
“In general, we are not formed with user data. So using it, or any query you put in, any answer you put in, we don't train the models with it,” he said, adding, “We ask for a lot of feedback from our users.”
In addition to the customization option, users can listen in the background in Audio Overviews. This allows you to continue working in NotebookLM, check your sources, receive quotes, and explore relevant quotes while the audio plays in the background.
NotebookLM was initially launched as a project at Google's I/O developer conference last year and debuted for public access in the US in December. In June it expanded to markets such as India, the United Kingdom and more than 200 countries. Although the product initially had some traction within education and research use cases, companies and organizations only began testing it after Google expanded its support for more fonts and added new features.
Now, Google says more than 80,000 organizations use NotebookLM, which it sees as an opportunity to explore monetization. Hoping to build on this traction, the company launched the NotebookLM Enterprise pilot program on Thursday.
Companies can apply for the pilot and, if accepted, Google said, they will get early access to product features, training and email support.
Martin told TechCrunch that under the enterprise pilot, his team trains organizations interested in using NotebookLM on how other companies use it.
“Us [also] “If we want companies to tell us, these are the features we want to implement,” he said.
Basic availability and pricing for NotebookLM Enterprise will be announced later this year. However, Google has not yet revealed the exact timeline or any specific details about the pricing tiers.
NotebookLM currently receives 4.17 million monthly visits, of which 2.5 million come from desktop computers and 1.6 million from mobile devices, according to SimilarWeb.
Currently, the assistant does not have a dedicated mobile application and is available on all screens through its website. However, Martin told TechCrunch that the team is actively exploring a native mobile experience to expand NotebookLM's presence among smartphone users. It's also exploring more voices, languages, and controls for audio overviews.
Additionally, the team explored and prototyped different numbers of speakers, to go beyond the existing two speakers for AI audio discussions, although this is not likely to be available anytime soon as Martin said it was not the most requested feature by users. the users.
Last month, NotebookLM added YouTube videos and audio files as sources for generating summaries along with existing sources like Google Drive, URLs, PDFs, and text.
Martin said that NotebookLM considers PDF files and YouTube videos as the two main sources. The team also observed a “very high percentage” of users who listened to a basic audio description and used the chat. The next largest group are users who only use chat without generating a basic audio description.