This post is all about one of the trickiest parts of the ChatGPT interface: Custom Instructions.
People think this is easy, and I find they often fill in LOTS of stuff about their lives, their personal preferences, the company’s priorities—stuff that ends up confusing and tricking the model into giving unhelpful responses to day-to-day work.
The answer isn’t to avoid Custom Instructions (CI), it’s to learn what they are for, where they slot in the prompt priority stack (there is one!), and to learn how to write an excellent Custom Instruction that actually simplifies your life.
And yes, this post includes lots of examples of real CI prompts, a punch list to get started, and pitfalls to avoid. As a bonus, I also call out where to use Custom GPTs vs Custom Instructions, since those are frequently confused.
Net net, this should give you an actual guidepost to what to do with those mysterious CI boxes.