NotebookLM adds agents for better research

Google is turning NotebookLM into more than an intelligent document reader: it can now search for sources, run code, analyze data, and prepare complete reports through a conversation.

The update adds agent capabilities, meaning features that let the tool chain together multiple steps to complete a task. NotebookLM can help you gather information, compare it with your documents, and turn it into a ready-to-use result.

A chat with more analytical power

NotebookLM now runs on Gemini 3.5 and Antigravity, according to Google. The company promises more accurate and reliable responses, along with greater visibility into how the system reaches its results.

Each notebook also includes a secure cloud computer. In practice, NotebookLM can write and run code to perform calculations, clean data, or carry out more complex analyses without requiring you to program.

The system includes more than 100 selected software skills. In Google's internal tests against the previous version, the update achieved an average win rate of more than 65% across its five main evaluation areas. The improvement was especially notable in:

• Long-document analysis: 69.9% win rate.
• Advanced web research and source discovery: 78.2%.

These figures compare the new system with Google's previous version, not with every research tool available.

From a conversation to a report or presentation

NotebookLM also expands the formats it can generate. You can give it detailed instructions and ask it to turn the content of your sources into a specific result, which you can then edit and download.

The new options include:

• Charts and visualizations in PNG or SVG.
• Reports and documents in PDF, DOCX, Markdown, or plain text.
• Images in PNG, JPG, or GIF through Nano Banana.
• Structured data in CSV or JSON.
• Excel spreadsheets.
• PowerPoint presentations.

For example, you could upload sales data from several countries, ask NotebookLM to standardize the formats, run the analysis, and prepare a PDF with charts. It can also turn hard-to-read technical specifications into a simple guide, a presentation, and a roadmap for a team.

You no longer need to start with all your sources

Until now, NotebookLM worked best when you arrived with your documents already prepared. The update lets you start with a vague idea or a question and ask it to help you build a research base.

It can search for relevant sources on Google Search, find related papers, or locate primary documents in other languages. You still decide which sources ultimately enter the notebook, and each response preserves attribution for the information it uses.

That matters because it limits one of AI's common risks: getting an apparently convincing answer without knowing where it came from. In NotebookLM, the goal is to let you review and control the material supporting the result.

What changes for you

The tool is no longer just a place to ask questions about your PDFs. It can help you prepare a financial analysis, compare research, create educational materials, or decide whether an advertising campaign is worth expanding to another city.

The features are rolling out globally on the web for Google AI Ultra users and Workspace business customers with AI Expanded Access. Google plans to extend them to more users over time.

What matters is not that NotebookLM produces more formats, but that it now covers more of the research process: finding information, checking it, analyzing it, and presenting it. The final quality will still depend on the sources you add and how carefully you review its results.