NotebookLM is undoubtedly one of Google's most impressive implementations of generative AI technology, providing users with the ability to explore documents and notes through the advanced Gemini AI model. In a significant update last year, Google introduced the capability to generate audio overviews of source material within NotebookLM. Now, this innovative feature is making its way into search results as part of a testing phase.
Instead of the traditional approach of clicking links or reading text-based summaries, users can now experience a unique format where two simulated voices narrate the search results. This feature, however, is not yet widely available. It is currently housed in search labs, requiring users to manually enable it. Anyone interested in this new audio overview experience can opt in to join the test.
Upon participating in the test, users will notice an embedded audio player appearing within their Google search results. Interestingly, this player is positioned below the standard block of AI-generated text, typically found after the initial search results and the “People also ask” knowledge graph section.
Google has designed this audio feature to be user-activated; it does not generate audio automatically. Users must click the generate button to initiate the playback process. After just a few seconds, they’ll be presented with a conversational summary delivered by two AI voices, which encapsulates the essence of the search results.
The audio player also provides a list of sources that contributed to the summary, along with controls to speed up or slow down the playback. For those looking for practical applications, Google recommends trying out searches like “how do noise cancellation headphones work”, but users are finding that Audio Overviews are appearing across various search queries.
In a curious twist, users can even search for Google audio overviews to hear an overview explain what an Audio Overview is. This self-referential feature showcases a fascinating yet somewhat unsettling aspect of technological progression.
Initial assessments of the audio overviews indicate that they are generally accurate for straightforward topics. However, there is a heightened risk for inaccuracies, especially as the AI summarizes content from a broader array of sources compared to the more controlled environment of NotebookLM, where users provide a limited set of documents and websites. This inconsistency is a challenge we've observed with text AI Overviews as well.
The expansion of Google's audio overviews is not limited to search results. Since their inception in NotebookLM, these audio summaries have also made appearances in Google's Gemini Deep Research mode and Google Docs. Given the rapid advancements in AI features from Google, it is anticipated that Audio Overviews will proliferate even further in the near future.
As a final note, users should be prepared for the possibility that Audio Overviews will become a standard feature in search results, even for those who do not participate in the current testing phase. After all, the text-based AI Overviews underwent only a brief laboratory testing period before being integrated into the default search experience.
