Google on June 8, 2026 announced a comprehensive set of upgrades to NotebookLM, its AI-powered research platform, moving the product from a document-analysis tool toward an autonomous research agent capable of writing code, browsing the web, and producing downloadable reports - though the new capabilities are limited to paid subscribers at launch.

What changed and why it matters

Three years after its debut as an experimental product from Google Labs, NotebookLM is getting its most substantial technical overhaul to date. According to the announcement authored by Trond Wuellner, Director of Product Management at NotebookLM, and Usama Bin Shafqat, Software Engineer at NotebookLM, the platform is now running on Gemini 3.5alongside a new framework called Antigravity - the same agentic infrastructure that Google unveiled at its developer conference in May 2026.

The changes are substantial across three distinct areas: the chat interface, the output system, and the process for starting research projects. None of the three areas is cosmetic.

For digital marketing and research professionals who use AI tools for synthesis and planning, the upgrade marks a meaningful shift. NotebookLM was previously most effective when users arrived with organised source material already assembled. The June 8 update tries to remove that prerequisite.

The cloud computer and 100+ software skills

The most technically significant component of the announcement is the addition of a secure cloud computer to each notebook. According to Google, this gives NotebookLM the ability to write and execute code directly within the research environment - a capability that was absent from all previous versions.

The cloud computer is paired with a library of more than 100 curated software skills. These skills function as pre-built capabilities the system can invoke during research tasks, enabling it to perform operations like data transformation, statistical analysis, and file conversion without requiring the user to write any code themselves.

Why does this matter? Consider a research scenario involving tabular data from multiple countries with inconsistent formatting. Previously, a user would need to clean and reconcile that data externally before bringing it into a notebook. According to the announcement, the upgraded system can now handle that normalization internally - writing code, running it, and then proceeding to analysis within the same workflow.

Benchmark performance: what the numbers say

Google published internal evaluation data alongside the announcement. In side-by-side comparisons against the previous system, the upgraded NotebookLM achieved an average win rate of over 65 percent across five core evaluation dimensions - representing a 15 percentage point margin above parity.

Two specific capability areas showed stronger results. For large document analysis, the system achieved a win rate of 69.9 percent. For advanced web research and source discovery, it reached 78.2 percent. Google stated these figures came from evaluations against the prior baseline system.

The figures are Google's own internal assessments, not independently verified measurements. The evaluation methodology - including what constitutes a "win" in these comparisons and how evaluators were selected - was not described in detail in the public announcement. That context matters when interpreting what 65 percent or 78.2 percent actually represents in practical terms.

Output formats: from chat to downloadable files

The second major area of change is the output system. NotebookLM has historically produced content within the chat interface itself - text responses, summaries, and audio overviews. The June 8 update extends this to eleven distinct downloadable formats, generated via what Google calls the studio panel.

According to the announcement, the new formats include:

  • Data visualizations and charts (PNG and SVG)
  • Documents including PDF reports, DOCX files, Markdown, and plain text files
  • Images using Nano Banana (PNG, JPG, and GIF)
  • Structured data in CSV and JSON formats
  • Microsoft Excel spreadsheets (XLSX)
  • Microsoft PowerPoint presentations (PPTX)

The breadth here is notable. PDF reports with embedded charts and tables, budget spreadsheets, and student worksheets are each cited as examples in the announcement. Users can also edit outputs after generation. Google indicated it plans to add additional formats in the future without specifying which ones or a timeline.

The Nano Banana integration for image output is a continuation of a rollout that began in October 2025, when the image generation model - built on Gemini 2.5 Flash - expanded from the Gemini app into Google Search and NotebookLM. At the time, Google reported that more than 5 billion images had been generated using the model since its August 2025 debut. That underlying technology is now powering the image output capability inside the new output formats.

Starting from scratch: web research and source building

The third component of the update addresses how research projects begin. Previously, a user needed to supply their own sources before NotebookLM could do much useful work. The updated system can now accept loose questions and ideas as starting points.

According to the announcement, NotebookLM can guide a user through building a source repository directly within the chat. It can find primary sources in other languages, locate related works by a specific author, and use Google Search to surface and add relevant web sources to the notebook automatically.

Users retain control over which sources are added, and all sources continue to be attributed. This is a meaningful design choice. The system is not autonomously assembling a knowledge base without oversight - it suggests and adds sources that users can accept or reject.

The practical consequence is that NotebookLM becomes usable earlier in a research process. Someone exploring an unfamiliar topic who does not yet know which sources to trust can now start the conversation in NotebookLM rather than doing preliminary research elsewhere first.

Who gets access

The upgrades are rolling out globally via the web starting June 8, 2026. Access is limited to two groups: users with a Google AI Ultra subscription, and Workspace business customers with either AI Ultra Access or AI Expanded Access.

Free-tier users are not included in the initial rollout. Google stated it plans to expand access to others over time but did not provide a schedule.

This access structure is consistent with the pattern Google has established since at least December 2024, when it launched NotebookLM Plus as a premium tier targeting enterprise and educational customers. At the time, Steven Johnson, Editorial Director at Google Labs, noted that millions of users and tens of thousands of organizations were already using the platform.

The mobile app, which launched in May 2025, is not mentioned in the June 8 announcement as a delivery surface for the new agentic capabilities. The upgrade is web-only at launch.

The Antigravity connection

The mention of Antigravity in the announcement is technically significant. As PPC Land documented after Google I/O 2026, Antigravity is Google's agentic development platform - a system where AI agents, rather than individual users, do much of the orchestration. The platform was introduced in late 2025 and received a major update at I/O 2026 on May 19.

Deploying Antigravity inside NotebookLM is Google's clearest signal yet that the underlying architecture of the product has changed. The prior versions of NotebookLM operated within a retrieval-augmented generation framework - a model where the AI queries a fixed set of user-supplied sources and synthesizes responses from them. The Antigravity layer introduces the ability to act beyond those sources: running code, browsing the web, and invoking software skills.

The implication for professionals using NotebookLM for complex research is that the system can now chain together tasks that previously required manual handoffs. A data analyst working across multiple datasets with inconsistent formats, for example, could ask NotebookLM to normalize the data, conduct supplementary web research, write code to perform calculations, and produce a formatted PDF report - all within one session.

What this means for marketing and research workflows

For the marketing community, the most directly relevant applications are in research-heavy workflows. The new output formats - particularly PDF reports with charts, Excel spreadsheets, and PowerPoint presentations - reduce the gap between AI-assisted research and presentable deliverables.

The Deep Research capability that Google added in November 2025 already addressed part of this problem: it automated the process of synthesizing reports from hundreds of websites. The June 8 update adds the ability to format and package those findings into files suitable for distribution, presentation, or further editing.

Small business owners are cited explicitly in the announcement. According to Google, a gym owner running a media campaign could use the upgraded NotebookLM to analyze raw sales data against ad spend, calculate campaign return using the code execution capabilities, and produce outputs ready to inform expansion decisions. That workflow - raw data in, structured analysis out - is the kind of task that has typically required spreadsheet skills or a dedicated analyst.

The access limitation is the significant constraint here. The capabilities are compelling, but they are available only to subscribers who are already paying for Google's highest-tier AI access. Free-tier researchers and small businesses not on Workspace do not have a path to the new features at this point.

Context: three years of NotebookLM development

The June 8 announcement is the latest step in a development arc that began in May 2023 when Google introduced the product under the experimental name Project Tailwind. The platform shed its experimental label in October 2024 and introduced business-focused audio features alongside customizable controls the same month.

Since then, the pace of capability additions has accelerated. Video Overviews launched in July 2025. Language support expanded to over 80 languages in August 2025. The November 2025 Deep Research update automated multi-source synthesis. The mobile app arrived in May 2026. The June 8 upgrade layers the Gemini 3.5 model, Antigravity orchestration, and the cloud compute capability on top of that existing foundation.

Each of these additions has expanded the surface area of what NotebookLM can do - but the June 8 update is the first that fundamentally changes what kind of tool it is. Moving from a document-grounded Q&A system to an agent that writes and runs code, browses the web autonomously, and produces formatted files represents a different category of capability.

Timeline

Summary

Who: Google, through Trond Wuellner (Director of Product Management) and Usama Bin Shafqat (Software Engineer) of the NotebookLM team at Google Labs.

What: A comprehensive upgrade to NotebookLM that moves the platform to the Gemini 3.5 model and Antigravity framework, adds a secure cloud computer with more than 100 curated software skills enabling in-notebook code execution, introduces 11 new downloadable output formats including PDF reports, Excel spreadsheets, and PowerPoint presentations, and allows users to begin research from loose questions rather than pre-assembled sources. Internal evaluations cited a 65 percent average win rate over the prior system, with 78.2 percent in advanced web research and 69.9 percent in large document analysis.

When: Announced and rolling out on June 8, 2026.

Where: Available globally on the web at notebooklm.google for users with Google AI Ultra subscriptions and Workspace business customers with AI Ultra Access or AI Expanded Access. Free-tier users are not included at launch.

Why: The upgrade reflects Google's broader push to embed its Antigravity agentic infrastructure across its product suite following the I/O 2026 announcements. For NotebookLM specifically, the changes address the gap between AI-assisted research and usable, formatted outputs - a limitation that constrained adoption among professional and business users who need deliverables, not just analysis.