Summary of "How to Setup Claude Skills in Gemini + Google Notebooks!"

Overview

The video explains a step-by-step workflow for creating Anthropic “Claude skills” using Google Notebook LM (powered by Gemini). The goal is to have Claude follow structured instructions based on your provided sources, reducing hallucinations and saving time compared to building skills from scratch.

What “Claude skills” are

Skills are instruction “folders” for Claude.
Each skill contains a skill.md file that teaches Claude how to complete a specific task (framed like a recipe, cheat sheet, or coach).
When Claude runs, it retrieves the relevant skill and follows the instructions—intended to prevent guessing and “made-up” answers.

Core interaction: Gemini + Notebook LM → Claude skills

Notebook LM acts as a research and grounding layer:

You load sources (e.g., PDFs, websites, YouTube videos, documents).
Notebook LM answers questions based only on those sources to avoid fake or unsupported information.

The process produces:

A ready-to-paste skill.md generated by Notebook LM.
Claude then performs execution by producing the requested output in the correct format and tone.

The tutorial workflow (step-by-step)

Open Notebook LM at notebooklm.google.com and sign in.
Create a new notebook.
Add high-quality, focused sources (quality in, quality out).
- Example sources mentioned for landing pages: top guides, conversion copywriting articles, expert YouTube content, case studies, and official Claude skills documentation.
Research in Notebook LM by asking questions such as:
- “What are the top elements of a high-converting landing page?”
- “What’s the best structure for a skill.md file?”
- “How do I write a skill that teaches converting copy?”
Generate the skill.md by prompting Notebook LM to create a structured skill template.
- Example template sections include: headline, subheadline, value proposition, social proof, CTA, and FAQ.
Move the skill into Claude:
- Install/prepare Claude Code (if needed via Anthropic’s site).
- Locate the skills folder.
- Create a new skill subfolder (e.g., “Landing page converter”).
- Paste the generated skill.md and save.
Test the skill in Claude:
- Ask Claude to generate a landing page using the skill.
- If results are off, go back to Notebook LM to tweak the skill and add sources, then regenerate and re-import.
- This emphasizes an iteration loop: Gemini/Notebook LM research → Claude refinement/execution.

Pro moves / best practices

Organize sources by topic (avoid mixing unrelated material).
Use multiple notebooks (one notebook per skill) to keep things focused and easier to update.
Keep skills small and single-purpose (“one skill, one job”):
- e.g., separate skills for headlines vs. FAQ vs. email writing, then combine/stack them.
Refine skills with Claude after Notebook LM generates skill.md:
- Ask Claude to improve/tighten the file and add usable examples.
Version your skills (treat updates like mini software product releases).
Test under pressure:
- Try edge cases and difficult scenarios; the “best skills” are battle-tested, not just theoretically correct.
Human safety check:
- Even with grounding, the presenter recommends reviewing skill.md before using it—“AI is fast, but humans still need to be the boss.”

Outputs may be grounded in sources, but the workflow still calls for human oversight.

Use cases/examples mentioned

Landing page conversion skills (headline/subheadline/value prop/CTA/FAQ)
Welcome email writing for new members (including onboarding steps and what to expect)
Examples of other possible skills:
- cold emails, sales scripts, customer support replies, product descriptions, SEO content
- lead magnets, onboarding flows, YouTube scripts

Key claims / positioning

The workflow is positioned as:

“Stupid easy” and faster than building skills manually
A way to reduce hallucinations by grounding outputs in provided sources
A scalable approach to build a stack of task-specific micro-skills for business automation