Deep Research Notebooklm

Deep Research via NotebookLM

Research $ARGUMENTS deeply using the NotebookLM MCP server and deliver a structured research brief. Optionally generate studio artifacts (slides, audio podcasts, videos, infographics, reports, mind maps) from the research.

Prerequisites

• NotebookLM MCP server must be configured. Install via: nlm setup add claude-code
• If NotebookLM MCP tools are not available, tell the user to run the setup command and restart their session.

Research Workflow

Step 1: Define Scope

Determine the research type based on the user's request:

Tell the user what you plan to research and confirm the angle:

"I'll research [topic]. My angle: [specific focus]. I'll investigate: [2-3 specific questions]. Sound right, or should I adjust?"

Wait for confirmation before proceeding.

Step 2: Create NotebookLM Notebook

Use notebook_create to create a notebook named: Research: [Topic] - [YYYY-MM-DD]

Step 3: Add Context Sources

Use source_add to seed the notebook with relevant context:

• Add any URLs the user provides (articles, company pages, reports)
• Add any documents or files the user references
• Add text summaries of relevant background if no URLs are available
• If researching a company, add their website, LinkedIn, recent press

Step 4: Run Research

Use research_start with a well-crafted query based on the topic and context.

Mode selection:

• Default: "fast" (~60 seconds, ~10 sources) -- good for most queries
• Use "deep" only if the user explicitly asks for exhaustive research (can take 10+ minutes and may stall at 0 sources)

Tip: Run direct WebSearch calls in parallel with NotebookLM for faster initial data gathering while the research engine works.

Poll research_status until complete. Use the query parameter as fallback matching -- task IDs can change between research_start and research_status calls.

Step 5: Import Discovered Sources

Use research_import to bring discovered sources into the notebook for deeper analysis.

Step 6: Query for Insights

Use notebook_query to ask 3-5 targeted questions based on the research type:

1. Overview: "What are the key findings about [topic]?" 2. Opportunities: "What opportunities or gaps exist in this space?" 3. Actions: "What are the most actionable insights from this research?" 4. Risks: "What are the main risks, challenges, or counterarguments?" 5. Custom: A question specific to the research type (e.g., "Who are the top 5 competitors and how do they differentiate?" for competitive intel)

Step 7: Write Research Brief

Save the findings to a local file using the research brief template:

File path: research/[topic-slug]-[YYYY-MM-DD].md

Use the template from research-brief-template.md to structure the output. Create the research/ directory if it does not exist.

Step 8: Present Takeaways

After saving, present the user with:

• 3-5 headline findings (bullets, direct, no filler)
• 1-2 recommended actions connected to the user's stated goals
• Surprises or contrarian findings -- anything that challenges assumptions
• The file path where the full brief is saved
• The NotebookLM notebook URL so the user can explore sources directly

Step 9 (Optional): Generate Studio Artifacts

Ask the user: "Want me to generate any artifacts from this research? Options: slides, audio (podcast), video, infographic, report, mind map."

If yes, use studio_create with the notebook_id from Step 2.

Available artifact types and recommended settings:

Common params for all artifact types:

• language: Set to the user's preferred language (e.g., "en", "es", "pt")
• focus_prompt: A clear directive about what to emphasize in the artifact
• confirm: Must be true to proceed with generation

After creating an artifact: 1. Poll studio_status until completed (audio/video: 5-15 min; slides/infographics: 2-5 min) 2. Use download_artifact to save locally if needed 3. Provide the notebook URL so the user can access artifacts directly

Tips:

• audio with deep_dive format produces the best podcast-style analysis
• slide_deck with detailed_deck format works best for standalone reading; presenter_slides is better when accompanied by speaker notes
• Audio status may show "unknown" once completed -- check for audio_url presence instead of waiting for a "completed" status

Notes

• Fast mode is recommended as the default. Deep mode is powerful but can take 10+ minutes and occasionally stalls.
• Always confirm the research scope with the user before starting -- a well-scoped query produces dramatically better results.
• The research brief template ensures consistent, actionable output across all research types.

Additional Resources

• research-brief-template.md -- Template for structuring research brief output

Research Brief: [Topic]

Date: [YYYY-MM-DD] Type: [Market Research | Competitive Intel | Client Research | Trend Analysis | Proposal Research | Academic/Technical] NotebookLM Notebook: [notebook name and URL]

---

Executive Summary

[2-3 sentences: what was researched, why it matters, top-line finding]

---

Key Findings

1. [Finding Title]

[Detail with data points, statistics, or evidence where available]

2. [Finding Title]

[Detail with supporting evidence]

3. [Finding Title]

[Detail with supporting evidence]

---

Opportunities

• [Specific opportunity with context on why it matters]
• [Specific opportunity with potential impact]

---

Risks & Challenges

• [Risk or challenge with context and severity assessment]
• [Risk or challenge with mitigation considerations]

---

Recommended Actions

1. [Action] -- [Why this matters and suggested next step] 2. [Action] -- [Why this matters and suggested next step]

---

Sources

• [Source 1 -- title, URL, and relevance]
• [Source 2 -- title, URL, and relevance]
• [Source 3 -- title, URL, and relevance]

---

Generated via deep-research-notebooklm skill using NotebookLM MCP