---
name: notebooklm-query
description: "Query NotebookLM notebooks through Claude Code with improved query handling, follow-up chains, and source filtering. Requires the nlm CLI tool."
---

# NotebookLM Query

Query Google NotebookLM notebooks through Claude Code with smarter query handling than the default MCP/CLI experience.

## Prerequisites

- The `nlm` CLI tool must be installed and authenticated
- At least one NotebookLM notebook must exist with sources added

## When to Use

- When researching a topic you have NotebookLM notebooks about
- When you want to query your curated knowledge base from the terminal
- When you need to cross-reference information across multiple sources in a notebook

## Process

### Step 1: Identify the Notebook

List available notebooks:
```bash
nlm notebook list
```

Ask the user which notebook to query, or infer from context if the topic makes it obvious.

### Step 2: Run the Query

**Important:** Pass the user's query to NotebookLM as close to verbatim as possible. The backend responds better to natural language questions than to restructured or overly formal prompts.

```bash
nlm notebook query --notebook-id <id> --query "<user's question, verbatim>"
```

**Do NOT:**
- Rephrase the query into a more "structured" format
- Break a single question into multiple sub-queries upfront
- Add prompt engineering to the query string

**Do:**
- Pass the question as the user asked it
- Let NotebookLM's backend handle the interpretation
- Add context only if the user's question is ambiguous without it

### Step 3: Present Results

Format the response clearly:
- Lead with the direct answer
- Include relevant quotes from sources with attribution
- Note which sources contributed to the answer
- Flag any gaps or areas where the notebook's sources don't cover the question

### Step 4: Follow-Up Queries

After presenting results, offer follow-up options:
- "Want me to dig deeper on any of these points?"
- "Should I query from a different angle?"

For follow-up queries, chain them — use context from previous answers to refine the next question. This produces better results than isolated queries.

```bash
# Follow-up with context from previous answer
nlm notebook query --notebook-id <id> --query "<follow-up question incorporating previous context>"
```

## Source Filtering

If the user wants to query specific sources within a notebook:

```bash
# List sources in a notebook
nlm source list --notebook-id <id>

# Query can reference specific sources by mentioning them in the query text
nlm notebook query --notebook-id <id> --query "Based on <source name>, <question>"
```

Use source filtering when:
- The notebook has many sources and the user wants information from specific ones
- Results are too broad and need narrowing
- The user mentions a specific document, article, or source by name

## Multiple Notebooks

If the user's question spans multiple topics:
1. Identify which notebooks are relevant
2. Query each one separately
3. Synthesise the results into a unified answer
4. Note which notebook each piece of information came from

## Known Limitations

- **Query length**: Very long queries may be truncated. Keep queries focused.
- **Source freshness**: NotebookLM sources are point-in-time snapshots. If the user asks about recent events, note that the sources may not be current.
- **Response length**: Some answers may be truncated. If a response seems incomplete, run a follow-up query targeting the specific missing piece.
- **Rate limits**: The API may rate-limit rapid successive queries. If you hit a limit, wait a moment before retrying.
- **Authentication**: If queries fail, the user may need to re-authenticate with `nlm auth`.

## Tips for Better Results

1. **Verbatim queries get better answers** — the backend is tuned for natural language
2. **Follow-up chains beat shotgun queries** — one good question followed by refinement beats five questions at once
3. **Name sources when you can** — "What does the McKinsey report say about..." beats "What do the sources say about..."
4. **Ask for comparisons** — "How does source A's view on X differ from source B's?" works well