notebooklm

What it does

Queries your Google NotebookLM notebooks directly and returns answers grounded in your source material — every claim cites the document it came from. This dramatically reduces hallucination because the AI is restricted to your notebook contents.

NotebookLM is Google's research tool where you upload sources (PDFs, articles, transcripts, etc.) and Gemini answers questions using only those sources.

When to use it

• Source-grounded research — you have a folder of clinical studies / interview transcripts / brand docs in NotebookLM and want a citation-backed answer
• Compliance questions — "what claims is the FDA okay with for ingredient X" (asked against your compliance docs)
• Brand voice questions — "how does this brand describe their target customer" (against brand strategy docs)
• Fact-checking — "did the interview mention X" (against transcripts)

Use this instead of general web search when you need answers grounded in specific documents you control, not the open internet.

Where it fits in the pipeline

flowchart LR
    A[Brief or<br/>compliance question] --> B[notebooklm skill]
    B --> C[Query a specific<br/>NotebookLM notebook]
    C --> D[Answer with<br/>source citations]

How to invoke

You: query my "Brand XYZ Research" notebook — what's their preferred
     tone of voice for older audiences?

Claude: [opens notebook in headless browser, queries Gemini against
         the notebook's sources, returns answer with citations]

Trigger phrases:

• "query my [notebook name]"
• "ask NotebookLM about [topic]"
• "check the [notebook name] notebook for [question]"

What you get back

• A short answer to your question
• Citations for every claim — which source document it came from
• Optional: full quoted passages from the source

Example:

Q: How does Brand XYZ describe their target audience?

A: Brand XYZ describes their target audience as "women in their 40s
   and 50s navigating perimenopause and menopause" [Source 1, p. 4].
   Specifically, they focus on women experiencing "brain fog, mood
   swings, and disrupted sleep" [Source 2, p. 12]. The brand
   positions itself as "the science-backed alternative to traditional
   HRT" [Source 1, p. 7].

Sources:
  1. brand-xyz-positioning-2026.pdf
  2. brand-xyz-customer-interviews.pdf

Setup

One-time setup (handled separately):

• Have a Google NotebookLM account
• Add your source documents (PDFs, articles, etc.) to a notebook
• The skill uses browser automation to connect; auth is persistent after first run

Example session

You: query my "magnesium-research" notebook — what's the evidence
     on glycinate vs citrate for sleep?

Claude: Querying "magnesium-research" notebook...

        Glycinate appears to have stronger sleep-quality evidence:
        - Two RCTs found glycinate improved sleep onset latency
          by 17-22 minutes vs placebo [Source 1, Souza et al. 2021]
        - Citrate showed inconsistent results — one trial found
          improvement, two found no effect [Source 3, Boyle et al. 2017]
        - Glycine itself is a known sleep modulator (separate from
          magnesium) [Source 2, Bannai & Kawai 2012]

        Caveat: the citrate trials used lower doses (200-300mg vs
        glycinate trials at 400-500mg), so dosing may be a confound.

        Sources:
        1. Souza-2021-mg-glycinate-sleep-rct.pdf
        2. Bannai-Kawai-2012-glycine-sleep-review.pdf
        3. Boyle-2017-magnesium-sleep-meta-analysis.pdf

Limitations

• Only queries the notebooks you've set up — doesn't search the open web
• Can only cite what's in the sources; if your notebook doesn't have an answer, it'll say so (which is actually the point — no hallucination)
• Notebook content has to be added by you upfront (NotebookLM doesn't auto-ingest)
• Doesn't accept Google Drive links as URL sources — paste the URL, NotebookLM grabs the Drive preview HTML, not the file. Use native file upload instead.
• Including the filename in your question can help when querying about a specific video/audio source