How to Use Google NotebookLM for Data Analysis

What NotebookLM Actually Does for Data Analysis

Google NotebookLM is a free AI research tool that grounds every response in sources you provide. Unlike general-purpose chatbots that draw on training data, NotebookLM only references the documents, spreadsheets, and files you upload. That constraint is precisely what makes it useful for data analysis: every insight, summary, and table it produces traces back to your data, with clickable citations.

Most guides focus on NotebookLM's audio overview and document summarization features. This tutorial covers a different workflow: using NotebookLM as a lightweight data analysis environment where you upload structured data, ask analytical questions, generate tables, and combine your datasets with web research for richer context.

Step 1: Create a Notebook and Upload Your Data

Go to notebooklm.google.com and sign in with your Google account. Click "New Notebook" and give it a descriptive name like "Q1 Sales Analysis" or "Survey Results 2026."

Add your data sources using the "Add Source" panel. NotebookLM accepts:

• CSV and TXT files (uploaded directly)
• Google Sheets (linked from Drive)
• PDFs with tabular data (financial reports, exports from BI tools)
• Google Docs containing data summaries
• Images of charts or tables (PNG, JPG, WEBP)

Each source can hold up to one million tokens, roughly 500,000 words, so even large datasets and multi-page reports fit comfortably. You can add up to 50 sources on the free tier or 300 on Pro.

Practical tip: If your data lives in Excel, export it as CSV before uploading. For Google Sheets, link directly so updates propagate automatically.

Step 2: Ask Analytical Questions in the Chat

Once your sources are loaded, the chat interface becomes your query tool. NotebookLM reads across all uploaded sources simultaneously, so you can ask questions that span multiple datasets.

Start with broad exploratory questions, then narrow down:

• "What are the top 5 products by revenue in this dataset?"
• "Summarize the key trends across all three quarterly reports."
• "Compare customer satisfaction scores between Region A and Region B."
• "Which months show the highest churn rate, and what patterns do you see?"

Every response includes inline citations. Click any citation to jump directly to the relevant section of your source data. This audit trail is critical for analysis work where you need to verify claims before acting on them.

What works well: Pattern recognition across documents, summarizing large tables, identifying outliers, and cross-referencing data points from different sources.

Where it struggles: Complex statistical calculations (standard deviations, regressions, hypothesis tests). NotebookLM is a reasoning tool, not a statistical engine. For heavy math, export your findings and run them through a dedicated tool.

Step 3: Generate Data Tables

The Data Tables feature in NotebookLM's Studio panel transforms unstructured or semi-structured information into clean, organized tables. This is particularly useful when your source data is embedded in PDFs or reports rather than already in spreadsheet format.

To generate a data table:

1. Open the Studio panel (right sidebar).
2. Select "Data Table" as the output type.
3. Specify the fields you want extracted. For example: "Create a table with columns for Product Name, Q1 Revenue, Q2 Revenue, and Year-over-Year Growth."
4. NotebookLM scans your sources and populates the table with matching data.

You can export these tables to Google Sheets with one click for further manipulation, pivot tables, or visualization.

Use case example: You have three PDF reports from different business units. Each contains revenue figures buried in paragraphs of text. Instead of manually extracting numbers, ask NotebookLM to build a consolidated comparison table across all three reports.

Step 4: Set Up Custom Instructions for Better Analysis

Custom Instructions let you configure how NotebookLM responds across your entire notebook. Access them through the gear icon in notebook settings. You get up to 10,000 characters to define your analytical framework.

Here is a custom instruction template optimized for data analysis work:

"You are a senior data analyst. When answering questions: (1) Always include specific numbers and percentages from the source data. (2) Flag any data quality issues you notice, such as missing values, inconsistent formats, or potential outliers. (3) When comparing metrics across time periods or segments, calculate the percentage change. (4) Distinguish between correlation and causation. (5) If the data is insufficient to answer a question, say so explicitly rather than speculating."

This instruction persists across all conversations in the notebook, so every query benefits from the analytical rigor without you having to re-prompt each time.

Step 5: Use Deep Research to Add Context

NotebookLM's Deep Research feature lets you supplement your own data with web research, all within the same notebook. Access it through the "Add Source" panel and select "Web," then choose between Fast Research (seconds, good for quick fact checks) and Deep Research (minutes, comprehensive multi-site analysis).

When to use it for data analysis:

• Your sales data shows a spike in March, and you want to know if a market event caused it.
• You are analyzing competitor pricing and need current public data to compare against your internal numbers.
• Your survey results reference industry benchmarks you do not have locally.

Deep Research browses hundreds of websites, compiles a report with citations, and lets you import the findings as a new source in your notebook. From that point forward, you can query across both your internal data and the research findings in a single conversation.

Step 6: Query Across Multiple Notebooks with Gemini

For analysis work that spans multiple projects or datasets, NotebookLM integrates with the Gemini app. This lets you mount several notebooks as data sources in a single Gemini conversation.

To set it up:

1. Open the Gemini app (gemini.google.com).
2. Click the "+" button in the chat input.
3. Select the NotebookLM notebooks you want to query together.
4. Ask cross-notebook questions like "Based on my customer feedback notebook and my product metrics notebook, which features drive the highest satisfaction scores?"

This is especially useful for quarterly reviews, board prep, or any situation where insights live in separate notebooks that were created at different times.

Limitations to Know Before You Start

NotebookLM is a reasoning and synthesis tool, not a replacement for Excel, Python, or dedicated BI platforms. Keep these constraints in mind:

Numerical precision. It can summarize and compare numbers from your sources, but it does not run SQL queries or statistical functions natively. For precise calculations, export data to Google Sheets.

Source count limits. The free tier caps you at 50 sources per notebook and 50 daily chat queries. Pro ($19.99/month) raises those to 300 sources and 500 queries.

No live data connections. Apart from Google Sheets links, there are no database connectors or API integrations. You work with snapshots of your data at the time of upload.

Privacy considerations. Your uploaded data is processed by Google's AI models. Review Google's data handling policies if you are working with sensitive or regulated data.

A Practical Workflow: Monthly Sales Review

Here is a concrete example of how this all fits together:

1. Upload your monthly sales CSV and last month's report PDF.
2. Set custom instructions for analytical rigor (Step 4).
3. Ask: "Compare this month's revenue to last month, broken down by region."
4. Generate a data table with columns for Region, Current Month, Previous Month, and Percentage Change.
5. Run Deep Research: "What macroeconomic factors affected retail spending in March 2026?"
6. Ask a final synthesis question: "Based on our sales data and the economic research, what drove the revenue increase in the West region?"
7. Export the data table to Google Sheets for your team presentation.

The entire workflow takes roughly 15 to 20 minutes, compared to the hours you might spend manually pulling numbers from PDFs and searching for context separately. If you want to skip the manual upload step entirely and get end-to-end analysis from a single prompt, tools like VSLZ handle the full pipeline from file upload to statistical analysis and charts without requiring you to configure notebooks or custom instructions.

What to Do Next

Start with a small dataset you know well, something where you can verify NotebookLM's outputs against your own understanding. A single CSV and one PDF report is enough to test the full workflow. Once you are comfortable with the chat query and data table features, experiment with custom instructions and deep research to see how much richer your analysis becomes.