3
2 Comments

I built a library of 2000 prompts for ChatGPT and Gemini

Hi everyone,

Over the past few months I’ve been experimenting a lot with AI tools like ChatGPT and Gemini. One thing I noticed quickly was that getting good results often depends on writing the right prompt.

I started saving prompts that consistently produced useful outputs. At first it was just a small list for my own use, but over time it grew into a much larger collection.

Eventually I ended up with around 2000 prompts across 20 categories, including things like writing, coding, productivity, business, research, and more.

To make them easier to use, I organized everything into a simple spreadsheet with separate categories, so it’s easy to browse and copy prompts when needed.

The goal was to create something practical that helps people save time and get better results from AI tools.

I’m curious to hear from others here:

What type of prompts do you use most often with AI tools?


If anyone is interested, you can check it out here:

‎https://dfyofficial.gumroad.com/l/2000prompts

posted toAvatar for product DFY - Done For You
DFY - Done For You
  1. 1

    Nice collection. To your question about what prompts I use most: I stopped using full prompts altogether. I break them into parts instead.

    The thing I kept running into with prompt libraries is that pre-written prompts are hard to adapt. You copy one, then spend 10 minutes editing it because the role is wrong, the output format doesn't match, or the constraints are too generic for your use case. The prompt looks useful but it was written for someone else's context.

    What works better for me is building prompts from reusable components. Define the role once, swap in different constraints, change the output format. Think of it like composable blocks instead of monolithic templates.

    That's actually what I built with flompt (flompt.dev). It's a visual canvas where you place typed blocks (role, objective, constraints, output format, examples, 12 types) and wire them together. Then it compiles to XML that Claude and other models parse cleanly. Your 2000 prompts would be interesting to decompose into blocks to see which patterns repeat across categories. Open source: https://github.com/Nyrok/flompt

  2. 1

    You should use scaffoldtool to redirect the prompts directly to ChatGPT.