Exploring Best Practices for Writing Instructions for GPTs

Has anyone else found it tricky to find guidance on writing GPT instructions, especially for actions’ interactions? Let’s share our experiences and learn from each other. What have you tried? What worked and what didn’t?


I am looking to understand the best way to write custom GPT instructions. When writing instructions in the configure tab, is it better to:

  1. Refer to the GPT directly as “you”? Ex: “You are an expert geologist. I will upload an image of a rock…etc”
  2. Refer to the GPT in the third person. Ex: “This GPT is an expert geologist. When the user uploads an image of a rock…etc”

When I use the create tab, it seems to write the instructions similar to example #2, but I find that it is difficult to write complex instructions clearly in this way. I’ve tried a few different approaches on my custom GPTs and I seem to be getting better outputs when I use method #1.

Does anyone have any insights (or resources) on how to frame instructions for custom GPTs?