Google NotebookLM Adds Expanded Thinking Steps and New Output Formats

Google updated NotebookLM to improve accuracy and reliability by switching its underlying AI models. The application now runs on Gemini 3.5 and Antigravity instead of previous versions. This change targets better performance across core evaluation dimensions.

Expanded thinking steps in chat improve transparency while new output formats support diverse data exports

NotebookLM adds expanded thinking steps in the chat interface to increase transparency into the model's reasoning process. Users can now see how the system processes queries before generating final answers. Google introduced this feature because users requested more visibility into the AI workflow for a long time.

The updated models achieved an average win rate of over 65 percent across five core evaluation dimensions. NotebookLM reached a 69.9 percent win rate when analyzing large documents. The system also delivered exceptional performance in advanced web research and source discovery with a 78.2 percent win rate.

NotebookLM now supports additional output formats for data visualization, documents, images, and structured files. Users can export results as PNG or SVG charts, PDFs, docx files, markdown, text files, JPG images, GIF animations, JSON, CSV, XLSX spreadsheets, and PPTX presentations. The application also enables multilingual input and output for global users.

Google rolled out these features globally to all eligible users. NotebookLM updates are available to subscribers with Google AI Ultra access. Workspace business customers who have AI Ultra access can also use the new capabilities immediately.

The update positions NotebookLM as a more transparent tool for document analysis and web research. The expanded thinking steps address previous user requests for better visibility into the AI process. The model upgrade directly supports higher accuracy in complex tasks.