NotebookLM Upgraded to Gemini 3.5: New Skills, Code Execution, and Output Formats
Google upgraded NotebookLM with Gemini 3.5, 100+ skills, code execution, and new output formats including PDF, PPTX, and CSV. Here's what changed.
What Google Just Changed About NotebookLM
Google’s NotebookLM has been a standout product since its launch — a research assistant that actually reads your documents and reasons over them rather than just returning search results. But the latest upgrade, powered by Gemini 2.5, is a meaningful step up from what it was.
This isn’t a minor model refresh. The update introduces code execution, over 100 new skills, and structured output formats including PDF, PPTX, and CSV — moving NotebookLM closer to a full productivity tool than a read-and-summarize assistant.
If you’ve been using NotebookLM for note-taking, research, or content work, here’s what’s actually new and what it means for how you work with it.
What Is NotebookLM, and Why This Upgrade Matters
NotebookLM is Google’s AI-powered research and note-taking tool. You upload sources — PDFs, Google Docs, web links, YouTube videos, audio files — and the assistant reasons over that specific content. It doesn’t pull from the general internet by default. It works from what you give it.
That scoped-context approach made NotebookLM useful for research, studying, and document analysis. But earlier versions had clear limits: the outputs were mostly text, the capabilities were relatively narrow, and there wasn’t much you could do with the results beyond copy-pasting them.
The Gemini 2.5 upgrade changes that in three concrete ways:
- More capable reasoning — Gemini 2.5 is substantially better at multi-step analysis, long-context understanding, and nuanced synthesis
- Expanded skill set — 100+ new capabilities the model can perform on your content
- New output types — You can now export structured files, not just text summaries
The combination makes NotebookLM significantly more useful for real work tasks, not just reading comprehension.
Gemini 2.5 and What Better Reasoning Actually Means
The underlying model upgrade from Gemini 1.5 to Gemini 2.5 isn’t just a benchmark improvement — it shows up in how NotebookLM handles complex requests.
Longer Context, Better Synthesis
Gemini 2.5 supports a massive context window. In practice, this means NotebookLM can hold more of your uploaded material in working memory at once, and draw connections across sources more accurately. If you’ve uploaded a 200-page report alongside three supplementary papers, it handles that more gracefully than before.
More Reliable Multi-Step Reasoning
Where earlier versions sometimes lost the thread on complex questions — especially ones requiring comparison, contradiction-finding, or structured analysis — Gemini 2.5 handles these better. Ask it to compare the methodology sections across five papers, or identify where two sources directly contradict each other, and the output is more coherent and accurate.
Reduced Hallucination on Grounded Content
Because NotebookLM works from your sources rather than general training data, hallucination risk is already lower than general-purpose chatbots. Gemini 2.5’s improved grounding makes this even more reliable. It’s less likely to invent citations or attribute claims incorrectly to a specific source.
The 100+ New Skills: What You Can Now Ask NotebookLM to Do
The “skills” expansion is one of the more practically significant parts of this update. Skills are structured capabilities — specific types of analysis or transformation the model can apply to your content.
What Skills Are
Think of skills as named operations: summarize this as a timeline, extract all claims with supporting evidence, generate a quiz, create a glossary, write a briefing document. The previous version had some of this, but the expansion to 100+ covers a much broader range of use cases.
Skill Categories
The skills span several categories:
Research and analysis
- Extract key arguments with citations
- Identify methodological weaknesses in academic papers
- Compare positions across multiple sources
- Flag areas of uncertainty or conflicting evidence
Content creation
- Draft executive summaries from technical documents
- Write blog post outlines grounded in your sources
- Generate social media posts from research notes
- Create email drafts from meeting notes
Learning and comprehension
- Generate study guides and flashcards
- Create practice questions at different difficulty levels
- Explain concepts in plain language
- Build glossaries from technical documents
Structured outputs
- Format findings into tables
- Create action item lists from meeting transcripts
- Generate structured briefings for different audiences
The breadth here is meaningful. NotebookLM is no longer just a “summarize this document” tool — it can perform specific, repeatable operations on your content.
Code Execution: The Feature That Changes the Most
Code execution is the most technically significant addition to NotebookLM. This lets the model write and actually run code — not just generate it — to perform analysis on your content.
What Code Execution Enables
Before this update, if you asked NotebookLM to do any kind of quantitative analysis — calculate averages from a dataset, identify trends, do unit conversions, count occurrences — it would either refuse or generate code you’d have to copy into a separate environment.
With code execution, it can:
- Parse and calculate from numerical data in your uploaded files
- Generate charts and visualizations from spreadsheet content
- Perform statistical summaries on research data
- Convert and format data between structures
- Run logic checks on information in your documents
A Practical Example
Suppose you’ve uploaded a CSV of survey results and a research paper. You can now ask NotebookLM to calculate the mean and standard deviation of specific columns, compare those numbers to the benchmarks cited in the paper, and tell you whether the results are statistically notable. It runs the actual calculation, not a hypothetical one.
This is a significant expansion in what’s possible without leaving the tool.
What It Doesn’t Do
Code execution here is scoped to working with your notebook content. It’s not a general-purpose coding environment, and it doesn’t connect to external APIs or databases unless you’ve uploaded that data. The execution environment is sandboxed. But for data analysis within the context of your research, it’s a substantial new capability.
New Output Formats: PDF, PPTX, and CSV
The output format expansion is arguably the most immediately practical change for knowledge workers. NotebookLM now lets you export its outputs as:
- PDF — polished documents for sharing or archiving
- PPTX — presentation slides ready to open in PowerPoint or Google Slides
- CSV — structured data tables for use in spreadsheets or downstream tools
Why This Matters
Previously, everything NotebookLM generated stayed in the tool as text you’d have to manually copy, format, and transfer elsewhere. That friction was a real barrier to using it in actual workflows.
Now you can generate a briefing document and export it as a formatted PDF. Create a presentation outline grounded in your research sources and get a PPTX you can open and edit. Extract a structured table of data points and download it as a CSV for analysis.
Specific Use Cases
PDF exports are useful for:
- Research summaries to share with colleagues
- Client-facing documents based on uploaded project files
- Formatted reports from document collections
PPTX exports are useful for:
- Turning research into presentation drafts
- Building slide decks from meeting notes or strategy documents
- Creating educational presentations from course materials
CSV exports are useful for:
- Extracting structured data from unstructured documents
- Compiling information from multiple sources into a single table
- Feeding NotebookLM analysis into spreadsheet tools
The net effect is that NotebookLM can now sit more naturally in the middle of a workflow rather than at the end of one.
How These Features Work Together
The individual upgrades are useful in isolation, but they’re more powerful combined.
Consider a researcher working on a competitive analysis:
- Upload 10 industry reports, three competitor websites saved as PDFs, and a spreadsheet of market data
- Use a skill to extract key claims from each report with citations
- Run code execution to calculate trend lines from the market data
- Ask for a synthesis that connects the quantitative trends to the qualitative claims
- Export the whole thing as a PPTX for a stakeholder presentation, and a CSV of the extracted data points for further analysis
One coffee. One working app.
You bring the idea. Remy manages the project.
That’s a workflow that previously required multiple tools, significant manual assembly, and probably several hours. With the updated NotebookLM, it’s one environment.
Where MindStudio Fits Into AI-Powered Knowledge Work
NotebookLM’s upgrade makes it more capable as a standalone tool. But for teams who want to build repeatable, automated workflows around AI — not just use it interactively — MindStudio offers something different.
MindStudio is a no-code platform for building AI agents and automated workflows. Where NotebookLM is a workspace you interact with manually, MindStudio lets you encode that kind of reasoning into agents that run on their own.
A few examples of how teams use it alongside tools like NotebookLM:
- Automated research digests: An agent that monitors new content, summarizes it, and delivers a structured briefing to Slack or email on a schedule
- Document processing pipelines: Agents that ingest uploaded files, extract structured data, and push it into Airtable or Google Sheets automatically
- Content workflows: Agents that take research inputs and produce drafts, formatted documents, or slide outlines — without manual prompting each time
MindStudio has 200+ AI models available, including Gemini 2.5, Claude, GPT-4o, and others — all without needing separate API keys. You can build an agent that uses Gemini for research synthesis, Claude for writing, and connects the outputs to your existing tools via 1,000+ pre-built integrations.
If you’re thinking about how to systematize the kind of work NotebookLM handles well, MindStudio is free to start and most agents take under an hour to build.
The distinction is simple: NotebookLM is great for interactive research sessions. MindStudio is better when you want that work to happen automatically, consistently, and at scale.
Frequently Asked Questions
What model does NotebookLM use now?
NotebookLM is now powered by Gemini 2.5 Pro. This replaces the earlier Gemini 1.5 Pro model. Gemini 2.5 Pro offers a larger context window, improved multi-step reasoning, and better performance on complex analytical tasks.
What are NotebookLM skills?
Skills are structured capabilities that NotebookLM can apply to your uploaded content. They range from research operations (extract claims, identify contradictions, compare sources) to content creation tasks (write summaries, generate outlines, create study guides). The latest update expanded the available skill set to over 100. Skills make NotebookLM more predictable and repeatable — you’re invoking a specific type of operation rather than just entering open-ended prompts.
Can NotebookLM run code on my data?
Yes, with the latest update. NotebookLM can now write and execute code on your uploaded content. This is most useful for quantitative analysis — calculating statistics, processing data in spreadsheets, generating visualizations. The execution is sandboxed and works on your notebook content, not external sources. It’s not a full development environment, but for data analysis within a research context, it covers most common needs.
What file formats can NotebookLM export now?
The new output formats are PDF, PPTX (PowerPoint-compatible), and CSV. PDF is useful for polished documents to share or archive. PPTX creates editable presentations from your research. CSV extracts structured data in a format compatible with spreadsheets and other tools. This is a significant change from earlier versions where all output was plain text you had to manually transfer and format elsewhere.
Is NotebookLM free to use?
NotebookLM has a free tier that covers core functionality. Google also offers NotebookLM Plus as a paid upgrade with higher usage limits, more notebooks, and access to premium features. The Gemini 2.5 upgrade and new features are available on both tiers, though some advanced capabilities may be restricted or rate-limited on the free version.
How is NotebookLM different from ChatGPT or Claude?
The primary difference is source grounding. NotebookLM works from documents you upload rather than general training data. When it answers a question, it references your specific sources. This makes it more reliable for research tasks and document analysis, and it can cite exactly where in your sources a claim comes from. General-purpose assistants like ChatGPT and Claude work from broad training knowledge and are better suited for open-ended generation tasks rather than source-grounded research.
Key Takeaways
- NotebookLM’s upgrade to Gemini 2.5 improves reasoning, long-context handling, and accuracy on complex analytical tasks
- The 100+ new skills make it possible to perform specific, repeatable operations on your content — not just open-ended chat
- Code execution is the most technically significant addition, enabling real data analysis and calculation within your notebooks
- PDF, PPTX, and CSV export formats let NotebookLM fit into real workflows rather than being a dead-end tool
- For teams who want these capabilities automated and running on a schedule, MindStudio lets you build AI agents that handle research, document processing, and content workflows without manual prompting — try it free to see how it compares
