Google rolled out a major update to its AI research tool NotebookLM on June 8, 2026 (US time). Alongside the familiar workflow of chatting with your own collected sources, you can now export results in 11 formats, including Excel and PowerPoint, and the underlying model has been refreshed to Gemini 3.5. Another significant change is that each notebook now has its own cloud environment capable of running code.

Eleven output formats let you export finished deliverables directly

The most practical part of this update is the greatly expanded set of output formats. Until now, NotebookLM returned its output as summaries, notes, or audio overviews, but you can now download what you have researched in 11 different formats.

The supported options include png and svg for data visualizations such as charts, document formats like PDF, Word (docx), Markdown, and plain text, images created with the Nano Banana image model in png, jpg, and gif, structured data in CSV and JSON, and Microsoft Excel (xlsx) and PowerPoint (pptx). Support for the two formats most often used in business, spreadsheets and presentation decks, is no small matter.

What is more, the conversion is not purely mechanical. You can attach detailed instructions when exporting. You can ask it to "format this into a report with this structure" or to "make an Excel sheet with these columns," so you receive your research in something close to the shape you actually want to use. In effect, the entire flow from gathering material to producing the final deliverable can now be handled in one stretch without switching between tools.

Gemini 3.5 and a "Secure Cloud Computer"

NotebookLM's brain has been swapped out as well. The underlying model is now Gemini 3.5, running on Antigravity, Google's agentic platform. Google says this improves the accuracy and reliability of answers while also making it easier to follow the reasoning process that led to a given conclusion.

In addition, each notebook is assigned its own Secure Cloud Computer. This is a cloud-based working environment where NotebookLM can write and execute code, reportedly with access to more than 100 software skills. The varied outputs mentioned above, such as Excel files, charts, and images, are possible precisely because this execution environment runs code behind the scenes to assemble the deliverables. The role of the tool has expanded from simply returning text to actively building finished outputs.

You can now start research from a vague question

The barrier to getting started has also been lowered. The earlier NotebookLM was at its best when users had already gathered their materials and had a clear idea of what they wanted to investigate. With this update, it can find and collect web-based sources from within the chat itself, so you can begin research even from a vague idea or question.

Because the tool can accompany you from the entry point of "I haven't yet figured out what to look into," it is easier to get a foothold at the very start of an investigation. Compared with the previous design, where the real work began only after the materials were in place, the range of situations where NotebookLM is useful has widened considerably.

Note that these new features are rolling out to paid tiers first. They are available to Google AI Ultra subscribers and to eligible Workspace business customers.

Summary

This NotebookLM update is built around three pillars: an expanded set of output formats, a refresh to Gemini 3.5, and the addition of a cloud environment that can run code. The ability to export to Excel and PowerPoint, and to shape deliverables by attaching instructions, is a practical advance that ties directly into everyday research and document creation. Although the rollout starts with paid tiers, it is worth watching as a sign that AI research tools are beginning to handle not just the "researching" but also the "putting it into shape" stage.