Creating Gems with Google Docs

Author's): Ime Inyang Jnr.

Originally published in Towards Artificial Intelligence.

In August 2024, Google released an interesting feature on the Gemini platform – Google's cutting-edge Multimodal Large Language Model (MLLM) – that allows users to customize models on the platform. It's called jewels (as you might guess, short for “Gemini”).

The Gems app was made available for free to subscribers in March 2025, making this tutorial useful for all parties.

In this tutorial, we'll show you how to create simple gems to suit different needs and how to make them more efficient using Google Docs.

Note: This tutorial assumes you already have a working Google account and are already using Gemini.

Creating a gem

Jewels

Gem is basically a Gemini model with a custom manual to aid user interaction. These statements are defined before you start interacting with the model. We call them System prompts.

System prompts allow language models to adopt personalities that match the user's needs. This eliminates the need to build a chat history with the model in order to build a certain level of proficiency in performing tasks.

These instructions are declared once and remain persistent in conversations.

Your first jewel

STEP 1: To create your first gem, visit Twinsopen the left panel with three parallel strokes and select “Discover Gems”.

This will take you to the gem manager where you will see your ready-made Google Gems and any personas you have created, if you have already done so.

STEP 2: Click the “New Gem” button. This will take you to the lab area where you can customize your gems.

STEP 3: Specify the name, description and instructions for your Gem.

Instructions are what Gems run on. This is where Prompt Engineering shines. Instructions can range from short tasks to a list of multimodal tasks. Great instructions are detailed and broken down.

STEP 4: Test your gem in the preview window on the right: interact with your gem and quickly see how it reacts after publishing.

STEP 5: Save your gem by clicking the “Save” button above the preview window.

Now you can go back to the Gemini home page and find your gem in the list.

Document based gems

Knowledge

To increase the power of your Gems, the lab has a feature called “Knowledge”. Knowledge is where you can provide your Gems with context to increase their capacity. These can be images, documents, code or slides.

You can even describe workflows in documents that your Gem will execute, as long as those processes are within Gemini's neutral capabilities.

Suppose you want to generate photos of a certain type, instead of always sending samples with every conversation, you can provide them in the knowledge area and instruct the Gem to follow the knowledge examples.

A career planning gem

Docs is my favorite program because of its editing flexibility and multimedia freedom. They can serve as a type of skill for your Gems.

I created a Career Planner using Docs. Here's how:

1. Create a new document in Drive.
2. Describe the name and function of the document: This is defined at the top of the document. Increases the Gem's confidence that they are using the correct knowledge. It also serves as a header for you, the builder, to know what document you are dealing with when you revisit it in the future.
3. Define the purpose of the document: This serves as a sort of extension of the Gem's instructions. Here I list what is expected of a Gem. If you expect a Gem to have a specific order, be sure to define it here.
4. Describe the output: A career planner is essentially a workflow: a series of steps and processes to carve a clear path. Each step has its expected inputs/outputs, which act as the source or resource for the next or previous step/process in the workflow. Here I define the expected data and result format for each step of the workflow.
5. Define your workflow: The workflow in this document is called “Questions” – it's a list of grouped questions for Gem to ask the user to understand their desired career direction.
6. Write in Markdown: To ensure that Gem understands the flow and the different levels at each stage, I wrote the entire process in a markdown format.
7. Add a document as knowledge: Return to your gem's lab area and select the “Add Files” button (+) under Knowledge and select “Add from disk”. Select the recently created document from the pop-up window. (You can define and add as many documents as your use case requires.)
8. Test: Interact with your Gem in the preview window and make sure it meets your expectations. You can edit the instructions or document until your Gem is working optimally.
9. Save: Once you are happy with the results of your Gem, you can save it and start planning your career path.

Markdown is a lightweight markup language with a simple syntax that uses plain text formatting to create structured documents. It basically uses hashtags (#) to identify headlines. The more hashtags (##), the lower the level of the headline.

This prompting strategy was adapted from the Anthropic strategy Skills functionality in Claude AI (see Here AND Here).

There's nothing that can't be done with a little quick engineering!

What makes this particularly useful is that everything is native to your Google account – no dragging and dropping required. Your gems are saved to your Google Drive.

You can also share your gems with others. Do this by:

• Going back to the Gem Manager and clicking the share button on the gem.
• Change general access from “Restricted” to “Anyone with the link”.
• Setting permissions for “browser”.
• Copy the link and click “Done”. This will notify you to grant view permissions for the document/knowledge source. Click “Share”.

Talk to A gem of a career plan

To get the most out of any knowledge, invoke it in the instruction window and specify how and when the knowledge should be used. This ensures that your Gem always refers to this knowledge when performing a specific task. Failure to do so may confuse the Gem, especially when providing a lot of knowledge.

You can experiment with how the Gem retrieves knowledge to ensure consistency in your learning, how much knowledge can be loaded at once, and the delay the Gem may experience when referencing its knowledge. This is important for the user experience.

Given that Gemini is multimodal, there is almost nothing that cannot be created or built into a Gem.

You could try building a slide creator/generator, a photo critique tool, or even a video creator. It all depends on your imagination, well-designed workflow and prompting skills.

Come in and have fun creating!

Attention:

Your Gem's functionality may be affected by your Google subscription plan, as higher plans provide enhanced Gemini capabilities (see details Here).

Published via Towards AI