NemoClaw Knowledge Wiki

Making and using Google Gems

3 min read

Making and using Google Gems

https://www.youtube.com/watch?v=Pwu0c5dUUOw https://www.youtube.com/watch?v=Pwu0c5dUUOw

Here is a detailed summary of the video by Tim Peakman regarding how to optimize workflow using Google Gemini Gems.

The Core Problem

Peakman argues that the majority of people use Large Language Models (LLMs) like ChatGPT, Claude, and Gemini incorrectly. They waste time in a “back-and-forth” chat loop trying to refine answers, often resulting in dissatisfaction. For solopreneurs and small teams, this inefficiency is costly. The solution is moving from generic chatting to Custom Assistants.

The Solution: Google Gemini Gems

“Gems” are Google’s version of Custom GPTs. They allow users to create specialized AI assistants pre-loaded with specific instructions and knowledge, drastically reducing the need for repetitive prompting.

The Framework for Creating a High-Quality Gem

Peakman emphasizes that AI output is only as good as the input. To create a successful Gem, he recommends following Google’s own four-part prompting formula found in their help section:

  1. Persona: Define who the AI is. (e.g., “You are an expert marketing copywriter and SEO specialist. Your tone is knowledgeable but informal.“)
  2. Task: Define exactly what the AI needs to do. (e.g., “Create a keyword-rich YouTube description designed to drive clicks.“)
  3. Context: Provide the background information necessary to complete the task. (e.g., “I will provide the transcript and title; you simply need to ask me for them.“)
  4. Format: Be specific about the structure of the output. (e.g., “Always start and end with the Call to Action link. Use bullet points for key learnings.“)

The “Knowledge” Feature: Crucially, Gemini Gems allows you to upload files (via Google Drive). Peakman suggests uploading Brand Guidelines and Brand Messaging documents so the AI inherently understands your business voice without needing to be reminded every time.

Practical Example: The “YouTube Description” Gem

Peakman demonstrates how he built a Gem specifically to write his weekly YouTube video descriptions. Step 1: Configuration

  • Name: YouTube Description.
  • Instructions: He pasted a prompt based on the 4-part formula above.
    • Constraint: He explicitly told the AI not to use salesy/hype language, make income guarantees, or mention competitors.
    • Formatting: He pasted a previous “perfect” description as a model for the AI to replicate.
  • Knowledge: He attached his specific brand documents.

Step 2: Execution

  • He opens the Gem, which is pre-programmed to ask him for four specific things:
    1. The Main Keyword.
    2. The Video Title.
    3. The Full Video Transcript.
    4. The Call-to-Action (CTA) Link.

Step 3: The Workflow

  • He inputs the keyword and title.
  • He downloads the subtitles/transcript from YouTube Studio for the specific video and pastes the entire text into Gemini.
  • He provides the CTA link.

The Result: The Gem analyzes the full transcript and produces a description that perfectly matches his required format (including timestamps, hashtags, and proper link placement) in seconds.

Key Takeaways

  • Systemization: This method turns repetitive tasks into a system, potentially saving 90% of the time usually spent on them.
  • Garbage In, Garbage Out: The quality of the Gem depends on the specificity of the constraints and the quality of the “Knowledge” documents uploaded.
  • “Starter for 10”: Peakman notes that the AI output should be viewed as a 90% complete draft (a “starter for 10”) that you review and tweak, rather than a final product to publish blindly.

Graph View