NotebookLM adds Audio Overviews, AI-generated podcast-style summaries

Google added Audio Overviews to NotebookLM, the research and note-taking tool it had launched as an experiment the previous year, letting users convert their uploaded sources — PDFs, Google Slides, web pages — into an audio discussion between two AI-voiced hosts summarising the material and drawing connections across it. Google said the feature drew on the multimodal capabilities of its Gemini 1.5 models.

Generation of a discussion for a large notebook could take several minutes, Google said, and the hosts spoke English only at launch; the company also acknowledged the AI hosts could make occasional factual errors and that listeners could not interrupt or redirect a conversation once generated. Despite the caveats, the format’s resemblance to an unscripted two-person podcast — filler words, back-and-forth banter, apparent surprise at material in the source documents — drew wide attention once users began sharing clips generated from unusual source material, well beyond NotebookLM’s existing base of students and researchers.

The feature marked a turn for NotebookLM from a document-summarisation tool toward a wider audience drawn by the audio format itself rather than the underlying research use case, and it prompted comparisons to and competition from a wave of AI podcast-generation tools that followed over the next year.