Google is testing podcast-like audio search summaries generated by AI. Audio Overviews uses Google’s latest Gemini models to generate “quick, conversational audio overviews for certain search queries.” It can be enabled through Google Labs, the company’s public-facing portal to AI experiments. An Audio Overview “can help you get a lay of the land, offering a convenient, hands-free way to absorb information,” Google says, noting that the feature displays search results “right within the audio player” to make it easy to delve further. Google already had AI audio summaries in NotebookLM and Gemini. Like those, Search features AI discussion “hosts.”

“When our systems determine it might be useful, you’ll see the option on the search results page to tap to generate a short audio overview on the topic of your query,” Google explains in a blog post.

The ensuing “AI podcast-style discussion” could produce a discussion about “something like, ‘How do noise cancellation headphones work?’” writes The Verge, explaining that once Google opts to provide Audio Overview results it will display “a button beneath the ‘People also ask’ module that says, ‘Generate Audio Overview.’” When clicked, the Audio Overview could take up to 40 seconds to generate.

“A few seconds later, you’re given a back-and-forth conversation between two AI voices summarizing the search results,” Ars Technica reports.

The audio player appears, embedded directly within Search, allowing you to adjust controls for things like mute and playback speed. Links to select Audio Overview sources appear below the playback bar.

Ars Technica calls NotebookLM “one of Google’s best implementations of generative AI technology, giving you the ability to explore documents and notes with a Gemini AI model,” and applauds the porting of Audio Overviews to Search.

However, a big difference between the Audio Overviews in NotebookLM and Search is that the former are “grounded by the small collection of documents and websites you feed into the system” while the audio Search results pull from the Web, Ars Technica points out, noting that this increases the possibility of errors.

While it calls the Audio Overview results produced in its test searches as “mostly accurate,” Ars Technica says it has experienced errors “repeatedly with text AI Overviews.”

Google is inviting users to opt in through Google Labs and leave feedback using a thumbs up/down on each discussion or on the experiment as a whole.