Google's NotebookLM attracted a loyal user base through AI assistance anchored in users' own content. The platform's latest capabilities give those users expanded tools for working with their notes.
On Monday, Google announced a wave of NotebookLM upgrades, with the biggest being an upgrade to Gemini 3.5 and Antigravity. As a result, users should experience more accurate results. Google said that internal side-by-side evaluations found that the upgraded NotebookLM achieved an average win rate of over 65% across the top five core evaluation dimensions against the prior system.
The upgraded system also equips each notebook with a secure cloud computer, which allows NotebookLM to write and run code to help users conduct deeper research, complex analysis and more than 100 curated software skills, according to the blog post. Ultimately, these upgrades are equipping NotebookLM to be a better-suited tool for needs that go beyond text.
Users can also ask NotebookLM to create outputs in more formats, now including: data visualizations and charts, documents (PDFs, docx, markdown, text files), images, structured data, and Microsoft Excel and PowerPoint.
Lastly, Google is evolving NotebookLM from a hub where users need to bring their own content for research help to one where users can get assistance in finding the information they need from a vast set of their own sources. NotebookLM now helps users build their own source repositories via the chat, where users can bring ideas or questions to get started.
It pulls from Google Search to find sources, and users can select what to add to their notebook, preserving NotebookLM's biggest appeal: control over which content is used to ground outputs. The updates are rolling out globally on the web to all users with Google AI Ultra and all Workspace business customers with AI Ultra access.
Our Deeper View #
Despite launching as a Google Labs experiment rather than a flagship product, NotebookLM has cultivated a remarkably loyal user base of millions, myself included. Its appeal is rooted in something more substantive than novelty: it directly confronts one of the most persistent concerns surrounding AI adoption, which is the risk of misinformation. Grounding responses in a user-defined dataset with web search disabled, it offers something rare in the AI space: genuine confidence in the sources behind your answers. This is a model more companies should recognize and build toward, and it's smart that Google is recognizing its value and building out its product offerings.