NotebookLM began with a focused proposition: turn the sources you provide into a grounded, interactive knowledge base. After uploading PDFs, websites, or notes, users could ask questions, summarize material, and turn it into podcast-style Audio Overviews.
Google’s June 2026 update pushes that source-grounded assistant much further. NotebookLM can now help discover sources, execute code for deeper analysis, and turn its findings into editable, downloadable files.
It is evolving from a reading tool into an end-to-end research workspace.
Gemini 3.5 and Antigravity under the hood
The upgraded NotebookLM uses Gemini 3.5 and Antigravity to improve reasoning on complex research tasks. Its chat experience offers greater visibility into the work being performed, helping users understand how the system searches, analyzes, and cites information instead of showing only a final response.
Google compared the new system with its previous version across five areas: accuracy and quality, multilingual support, large-document analysis, artifact creation, and web research. The upgraded system achieved an average win rate above 65%, a 15-percentage-point margin over parity.
Large-document analysis reached a 69.9% win rate, while advanced web research and source discovery reached 78.2%. The emphasis is not merely better conversational answers, but stronger performance across multi-document and multi-source research workflows.
A secure cloud computer for every notebook
The most consequential change is an isolated cloud computing environment attached to every notebook. NotebookLM can write and run code for tasks that text-only question answering cannot reliably complete, including:
- Cleaning inconsistent datasets;
- Combining tables from different regions or systems;
- Performing statistical and financial calculations;
- Creating charts from analysis results;
- Assembling the data, evidence, and conclusions into a report.
The system includes more than 100 curated software skills covering common research activities such as data processing, analysis, and visualization.
This materially changes the workflow. Previously, a user might move repeatedly between NotebookLM, Python, Excel, and presentation software. Now they can describe the desired outcome and let the system move from source comprehension through code execution to a finished deliverable.
A data analyst, for example, could provide several regional datasets with conflicting formats. NotebookLM could research the necessary context, write code to normalize the data, perform the analysis, and generate charts and a PDF report—all within the same notebook.
From chat responses to editable deliverables
NotebookLM previously focused on chat, reports, study guides, flashcards, quizzes, and audio or video overviews. The upgrade expands its downloadable output formats to include:
- Data visualizations and charts: PNG and SVG;
- Documents: PDF, DOCX, Markdown, and text;
- Images generated with Nano Banana: PNG, JPG, and GIF;
- Structured data: CSV and JSON;
- Microsoft Excel workbooks: XLSX;
- Microsoft PowerPoint presentations: PPTX.
Users can give detailed instructions about content, structure, and audience, then edit generated outputs from the Studio panel. NotebookLM is therefore no longer limited to explaining an answer; it can transform research into artifacts a team can actually use.
A program manager might convert a complex technical specification into a simplified guide, a slide deck, and a step-by-step implementation roadmap. A small-business owner could upload advertising and sales data, ask the system to calculate a campaign’s financial impact, and receive a decision-ready report.
Research can now begin without a prepared source library
NotebookLM has traditionally emphasized working from sources supplied by the user. That grounding helps reduce unsupported answers, but it also meant users had to collect their materials before meaningful work could begin.
The new experience lowers that barrier. A user can begin with a loose idea or question, and NotebookLM can help refine the direction, use Google Search to find relevant high-quality sources, and build a source repository within the chat. It can also discover primary material in other languages, bringing additional regions and perspectives into the project.
Users retain control over which suggested sources enter the notebook, and generated work continues to cite its evidence. NotebookLM is shifting from “bring me your sources and I will read them” to “let us assemble the evidence and analyze it together.”
What NotebookLM is becoming
This release connects retrieval-augmented generation, code execution, and artifact creation into one continuous workflow:
- Begin with a question or an existing source collection;
- Search for and select additional trustworthy sources;
- Perform source-grounded Q&A and cross-document reasoning;
- Write and execute code to process data;
- Generate charts, reports, spreadsheets, and presentations;
- Edit and download the results from Studio.
The distinction from a general-purpose chatbot is also becoming clearer. NotebookLM is not primarily designed to answer everything from model memory. It works around an inspectable source collection and keeps evidence, analysis, and deliverables organized within a single project space.
Code execution does not guarantee correct results. Users still need to inspect source quality, methodology, generated code, and conclusions—especially for financial, medical, or legal work. A cloud computer expands what the system can execute; it does not remove human responsibility for research judgment.
Availability
The features began rolling out globally on the web on June 8, 2026. Initial access covers Google AI Ultra subscribers and Workspace business customers with AI Ultra Access or AI Expanded Access. Google says availability will expand over time.
NotebookLM is no longer merely an AI that summarizes PDFs. By combining source discovery, evidence management, code-based analysis, and finished-file generation, it is becoming a source-centered research agent capable of doing substantive work.
