The secret to getting perfect answers from Gemini

By now, you've probably written quite a few commands for AI chatbots like Gemini or ChatGPT . With powerful large language models (LLMs) supporting these chatbots, Gemini or ChatGPT don't need much instruction to produce good results. However, the quality of the commands you provide to the AI ​​model directly correlates with the quality of the results you get. If you want to get better responses from your AI chatbot, you need to provide them with better commands – that's the science.

 

To improve the quality of your statements, you need to start with statement generation techniques, including optimizing and creating task-specific LLM inputs. Some long-time Gemini users have recently learned about a statement generation technique called meta-prompting. Essentially, this method uses AI to generate the final statements that will be fed into the chatbot. It's an easy way to improve your statement generation skills with virtually no manual effort, and if you haven't used it yet, you're falling behind.

What is meta-prompting?

General guidelines are used to create specific prompts.

 

If you're new to the world of prompt engineering, this Prompt Engineering Guide is a valuable resource as you learn the different strategies for writing successful AI input prompts. Let's start by explaining what meta-prompting actually is – it's a prompt-building technique that helps AI chatbots like Gemini understand how to approach a problem. Meta prompts add structure and syntax to guide the LLM powering the chatbot in the right direction.

You can write your own meta prompts, but a quicker, easier way to get started with meta-prompting is to have Gemini generate detailed inputs. This idea came from Anna Bortsova, a UX engineer at Google DeepMind, who uses Gemini to create super-level suggestions that are fed into Google 's AI tools like Veo 3. Bortsova's suggestions, created with Gemini, can be multi-page long, providing more detail and specificity than human-written guides.

"There are no rules here—we're experimenting—but I've found a few things that help Gemini generate really rich hints," Bortsova writes. "You need to define a very specific task: 'Write a detailed hint that an LLM specialist will understand.' And you need to be clear about your format and style. For example, an 8-second stop-motion animation of scenes rendered on paper. Then, introduce bindings, such as foil or glossy paper, instead of just plain paper. Then, let it do its own thing."

If anyone knows the best way to use Gemini, it's the engineers at Google DeepMind. So, use Gemini to create your own meta-prompting, and the results will be amazing.

Meta-prompting has changed the way people use AI.

Gemini helps users create images and build custom applications.

 

Meta-prompting with Gemini expands the possibilities for using AI applications, because detailed and optimized prompts become the standard. When using Google's Antigravity application to build custom Android and web applications like the one above, many people didn't write the prompt given to Antigravity. Instead, they gave Gemini the following prompt, asking it to write instructions for an agent-based integrated development environment (IDE):

Create a detailed prompt for an AI-based IDE that will build a simple play counter application for each song in this user's Discogs library: https://www.discogs.com/user/dippinthrutherecords

I need a simple interface in the web application so I can count how many times I've played each release and see the rankings of the ones I've played the most.

This only includes albums that I have in my Discogs collection.

Gemini returned a detailed prompt that was somewhat confusing. The meta-prompt generated by Gemini was extremely specific, defining the structure and syntax that Antigravity needed to use to respond to the user's request. It included in-depth programming knowledge to guide the AI ​​agent in the right direction. All of this came from the very simple prompt you see above, and the result was amazing.

This is the meta prompt Gemini provided for Antigravity:

# Role

You are a professional React/Next.js developer with experience integrating third-party APIs and managing local state.

Project Objectives

Build a personal web application called "Vinyl Spin Counter".

The application needs to retrieve a specific Discogs user's disc collection and provide a simple interface to track the number of times each disc has been played.

# Data Source

* **Target users:** `dippinthrutherecords`

* **Source URL:** https://www.discogs.com/user/dippinthrutherecords

**API:** Use the Discogs API (`GET /users/{username}/collection/folders/0/releases`) to retrieve the collection.

*Note:* Focus only on the standard "Collection" (folder ID 0). Do not retrieve the desired list.

Technology used

* **Framework:** Next.js (App Router).

* **Style:** Tailwind CSS.

* **Icon:** Lucide React (for the play button/rating icon).

* **Data Storage:** Use `localStorage` for MVP (Minimum Viable Product) to store play counts, ensuring data is persisted between page reloads.

* **State Management:** React Hooks (Zustand or the standard `useState`/`useEffect`).

Key Features

## 1. Collection Grid (Home View)

* Retrieve all releases from the `dippinthrutherecords` collection.

* Display them in a responsive grid (using Tailwind).

 

* **Card Layout:**

* Album cover image (Thumbnail).

* * Artist Name & Album Title (Abbreviate if too long).

* **Distinctive Counter User Interface:** A large number displays "Total Plays" with a large `+` button to increase the number of plays and a smaller `-` button to correct errors.

## 2. Leaderboard (Statistics View)

* A separate tab or pop-up window displays a ranking list of "Top 10 most played songs".

* Sort the data based on the highest number of plays stored locally.

## 3. Search & Filter

* A simple text search bar at the top allows you to filter the grid by Artist Name or Album Name instantly.

# Step-by-step implementation plan

1. **Setup & Data Types:** Initialize the project and define the TypeScript interface for `DiscogsRelease`.

2. **Client API:** Create a utility function to retrieve collection data. *Important: Handle pagination if the collection is large (Discogs paginates the results).*

3. **State Logic:** Create a custom hook `useSpinTracker` to:

* Load the number of plays from `localStorage` when mounted.

* Combine the number of plays with the retrieved Discogs data (using the release ID as the key).

* Display the functions `incrementSpin(id)` and `decrementSpin(id)`.

4. **User Interface Components:** Build the `ReleaseCard` and `Leaderboard` components.

5. **Assemble:** Assemble the main page.

Special Instructions

* **API Token:** If the Discogs API requires authentication for this public collection, allow me to define `DISCOGS_TOKEN` in the `.env.local` file and use it in the header.

* **Aesthetics:** Dark mode by default. Uses the "Vinyl Record" style (dark gray, circular pattern for the profile picture).

Let's start by setting up the project and defining the API data retrieval logic.

With just a few additional prompts, Antigravity was able to create a custom Vinyl Spin Counter application based on Gemini's instructions. The working and installable Android APK application demonstrated all the benefits of AI-powered prompting . The end result was better than simply providing Antigravity with a plain, human-written prompt. Furthermore, the AI-powered prompting strategy was more efficient, requiring fewer requests. This helped professional users avoid exceeding rate limits or wasting tokens too quickly.

While you can use AI chatbots to help write AI-powered prompts for complex tasks, they also work well for simpler ones. In the image gallery below, you can see the difference between images created using custom prompts and those created using Gemini's meta prompts:

 

There are very few instances where you wouldn't want to use an AI chatbot to write desired prompts on your behalf. There are virtually no downsides – it's often faster than meticulously writing prompts yourself, and including additional details will improve the quality of Gemini's output.

When should you use meta-prompting?

Whenever specificity is important, use AI to write suggestive questions.

By now, you're probably wondering when to use Gemini to generate meta prompts. Meta prompting can be useful for any LLM-related task, but it's most helpful for tasks and topics you're unfamiliar with. For example, when using Gemini to write a programming prompt for Google Antigravity, the meta prompt it generates includes computer science knowledge you would never have encountered.

Whether you're trying to create an image or build an application, meta prompting with Gemini can help you achieve better results.