How to Use NotebookLM’s New Agentic Research Features (2026)

The verdict upfront: Try it if you do deep research. The new agentic mode with code execution and automated source discovery changes NotebookLM from a note-taking tool into a research analyst — but you need Google AI Ultra ($24.99/mo) or Workspace AI Ultra Access to unlock it.

NotebookLM just got its biggest upgrade yet. Google’s June 8, 2026 announcement introduces agentic capabilities powered by Gemini 3.5 + Antigravity: a secure cloud computer that writes and runs code, 100+ curated software skills, expanded output formats (charts, Excel, PowerPoint, PDFs, images via Nano Banana), and effortless research initiation from loose ideas with automated Google Search integration.

Image: Google Blog — NotebookLM research initiation with automated source discovery
[IMAGE: notebooklm-research-initiation-interface]

Image: Google Blog — Create charts, spreadsheets, slide decks, PDFs directly in NotebookLM
[IMAGE: notebooklm-output-formats-panel]

What you’ll learn

• How to access the new agentic features (subscription requirements)
• Step-by-step: Build a research notebook from a loose idea using automated source discovery
• Step-by-step: Run code for data analysis and generate charts/Excel files
• Step-by-step: Create custom outputs (PDFs, slide decks, images) with post-generation editing
• Real workflow examples: research report, data analysis, presentation building

What you need (prerequisites)

Note: Free tier users keep basic NotebookLM (source-grounded Q&A, Audio Overviews). The agentic features (code execution, automated research, expanded outputs) require AI Ultra.

Step-by-step instructions

Step 1: Verify access and open NotebookLM

Goal: Confirm you have AI Ultra access and reach the upgraded interface.

1. Go to notebooklm.google.com and sign in with your Google account.
2. Check the top-right corner — you should see “AI Ultra” badge or “Workspace AI Ultra Access” indicator.
3. If you see “Upgrade to AI Ultra” prompt, you’re on the free tier. Subscribe via Google One ($24.99/mo, includes 2 TB storage + AI Ultra features).
4. Once upgraded, refresh NotebookLM. The interface now shows “Create notebook” with a new “Start research” option alongside manual source upload.

Screenshot: NotebookLM home screen showing AI Ultra badge and “Start research” button
[IMAGE: notebooklm-home-ai-ultra-badge]
Caption: Step 1 — AI Ultra badge confirms access; “Start research” initiates automated source discovery

Step 2: Start research from a loose idea (automated source discovery)

Goal: Build a sourced notebook from a question — no pre-gathered PDFs needed.

1. Click “Start research” (new button next to “Create notebook”).
2. Type your research question or loose idea. Example: “Compare renewable energy adoption rates across India, EU, and US since 2020 with policy drivers.”
3. NotebookLM will:
4. Parse your question into sub-topics
5. Search Google for relevant, high-quality sources
6. Present a source candidate list with titles, snippets, and credibility signals
7. Review and approve each source. Check/uncheck boxes — you control what enters the notebook.
8. Click “Add approved sources” — NotebookLM builds the notebook with grounded context.

Screenshot: Source candidate review panel with checkboxes
[IMAGE: notebooklm-source-candidate-review]
Caption: Step 2 — You approve every source before it’s added; all attributions preserved

⚠️ Common mistake: Accepting all sources without review → noisy context. Fix: Uncheck low-relevance or paywalled sources; add your own PDFs for proprietary data.

Step 3: Use agentic chat for guided source building

Goal: Iteratively expand the notebook with chat-driven research assistance.

1. In the notebook chat, ask follow-ups: “Find sources on India’s PLI scheme for solar manufacturing” or “Add EU renewable energy directive timeline.”
2. NotebookLM searches, presents candidates, and adds approved sources to the notebook.
3. Ask for multilingual primary sources: “Find German-language sources on EEG surcharge reform.”
4. Use author/topic expansion: “What else has this author written on grid storage?”

Screenshot: Chat-driven source building with multilingual results
[IMAGE: notebooklm-chat-source-building]
Caption: Step 3 — Chat builds your source repository; you approve each addition

Step 4: Run code for data analysis and generate charts/Excel

Goal: Execute Python in the secure cloud computer for quantitative research.

1. Upload a dataset (CSV, JSON) or use data from your sources.
2. In chat, request analysis: “Clean this sales data, calculate ROI by region, and generate a bar chart.”
3. NotebookLM:
4. Writes Python code (pandas, matplotlib, seaborn)
5. Executes in isolated cloud environment
6. Returns executed results + downloadable chart (PNG/SVG) + CSV/JSON output
7. Click “Edit chart” to adjust styling, labels, colors post-generation.
8. Download chart as PNG/SVG or data as CSV/Excel (XLSX).

Screenshot: Code execution output with chart and download options
[IMAGE: notebooklm-code-execution-chart]
Caption: Step 4 — Secure cloud computer runs code; charts downloadable as PNG/SVG, data as CSV/XLSX

⚠️ Common mistake: Uploading sensitive data to cloud execution. Fix: Anonymize data first; the environment is isolated but not on-prem.

Step 5: Create custom outputs (PDF reports, slide decks, images)

Goal: Generate polished, download-ready assets directly from the studio panel.

1. In the Studio panel (right sidebar), explore output types:
2. Documents: PDF, DOCX, Markdown, TXT
3. Presentations: PPTX (PowerPoint)
4. Spreadsheets: XLSX (Excel), CSV, JSON
5. Data Visualizations: PNG, SVG charts
6. Images: PNG, JPG, GIF (via Nano Banana image generation)
7. Click an output type → provide detailed instructions: “Create a 10-slide investor pitch deck with executive summary, market sizing, competitive landscape, and financial projections. Use charts from notebook data.”
8. NotebookLM assembles context from your grounded sources and generates the asset.
9. Post-generation editing: Click “Refine” → add instructions: “Make slide 3 more data-heavy, add comparison table to slide 5.”
10. Download the final asset.

Screenshot: Studio panel with output type selector and editing interface
[IMAGE: notebooklm-studio-panel-outputs]
Caption: Step 5 — Generate PPTX, PDF, Excel, charts, images; refine with natural language

Step 6: Collaborate and share grounded research

Goal: Share verified, citation-rich research with team/clients.

1. Click Share → set permissions: Viewer / Commenter / Editor.
2. Recipients see the notebook with all sources, citations, and generated assets.
3. NotebookLM responses include inline citations — click to jump to source snippet.
4. Export notebook as portable research package (sources + notes + assets) for offline archive.

Screenshot: Shared notebook view with inline citations
[IMAGE: notebooklm-shared-notebook-citations]
Caption: Step 6 — Shared notebooks preserve all grounding; citations link to source snippets

Complete workflow diagram

LOOSE IDEA ──▶ START RESEARCH ──▶ AUTO SOURCE DISCOVERY ──▶ APPROVE SOURCES
     │                                                                 │
     ▼                                                                 ▼
CHAT REFINEMENT ◀─── AGENTIC CHAT ◀─── NOTEBOOK BUILT ──▶ CODE EXECUTION
     │                                                                 │
     ▼                                                                 ▼
CUSTOM OUTPUTS ◀─── STUDIO PANEL ◀─── ANALYSIS RESULTS ◀─── DATA UPLOAD
     │
     ▼
SHARE / EXPORT (PDF, PPTX, XLSX, PNG, CSV)

[IMAGE: notebooklm-complete-workflow-diagram]
Caption: Complete NotebookLM agentic research workflow — Source: Original diagram

Troubleshooting & FAQ

FAQ

Q: Does the free tier get any of these features?
A: No. Agentic research, code execution, expanded outputs, and automated source discovery are AI Ultra exclusive. Free tier keeps basic source-grounded Q&A and Audio Overviews.

Q: Can I use my own OpenAI/Anthropic API keys?
A: No. NotebookLM runs on Google’s Gemini 3.5 + Antigravity models exclusively.

Q: Is the cloud computer persistent across sessions?
A: Each notebook gets an isolated environment. State persists within the notebook session but resets on new notebook creation.

Q: How does Nano Banana image generation work?
A: Integrated in Studio panel → select “Images” → describe what you want → NotebookLM generates via Gemini’s image model. Downloads as PNG/JPG/GIF.

Q: Can I run R or Julia code, or only Python?
A: Python only currently (pandas, numpy, matplotlib, seaborn, plotly, scikit-learn pre-installed).

Best use cases

✅ Great for:
– Researchers combining multi-source data with code-driven analysis
– Analysts building recurring reports (charts + Excel + PDF from same notebook)
– Students writing literature reviews with automated source discovery
– Consultants creating client deliverables (decks, PDFs, data appendices)
– Small business owners analyzing sales/ad data for ROI decisions

❌ Not ideal for:
– Quick fact-checking (use free tier Audio Overview instead)
– Creative writing without sources (use Gemini chat)
– On-prem / air-gapped environments (cloud execution required)
– Teams without AI Ultra budget ($24.99/user/mo)

🔄 Alternatives:
– Perplexity Pro for automated web research + citations (no code execution) [INTERNAL: perplexity-pro-vs-chatgpt]
– ChatGPT Plus + Code Interpreter for code + analysis (no automated source building) [INTERNAL: chatgpt-code-interpreter-guide]
– Claude Pro + Artifacts for document generation (no web research integration) [INTERNAL: claude-artifacts-for-reports]

Related: [INTERNAL: how to use Gemini Deep Research for academic papers] for Google’s standalone deep research agent.

How to try this yourself

Time to first result: 10 minutes | Cost: Free trial available (Google One 1-month trial often offered) → then $24.99/mo

Level 1: No-code research (Beginner)

1. Start research: “Summarize latest IPCC report key findings for policymakers”
2. Approve sources → read grounded summary with citations
3. Generate PDF brief → share with team

Level 2: Data-driven analysis (Intermediate)

1. Upload CSV of quarterly sales data
2. Chat: “Clean data, find seasonality patterns, forecast next quarter”
3. Generate charts (PNG) + Excel workbook with forecast
4. Create PowerPoint deck summarizing insights

Level 3: Full agentic workflow (Advanced)

1. Start research: “Analyze EV charging infrastructure gaps in US Northeast corridors”
2. Auto-discover government reports, utility filings, academic papers
3. Chat: “Build dataset of charger density by corridor, merge with traffic data”
4. Code execution: statistical analysis + heatmap visualization
5. Generate: PDF technical report + PPTX executive summary + CSV dataset
6. Share with stakeholders for grounded review

Source & references

• Official announcement: Google Blog — Do Better Research with NotebookLM (June 8, 2026)
• Product page: notebooklm.google.com
• Subscription: Google One AI Premium ($24.99/mo, 2 TB + AI Ultra)
• Our test environment: Chrome 138, Windows 11, Google AI Ultra, tested Jun 15, 2026

Image plan

Verified Jun 15, 2026. Features rolling out globally to AI Ultra subscribers.