The Universal Anatomy of the Prompt—How and Why a Prompt Works and How to Make it Work for You

I've written about prompting tips for individual models in the past but think of this as a level deeper—digging into the underlying structure that runs beneath good prompting, regardless of the model

I’ve written a fair bit about prompting tricks that work well with o3 (see here and here), but o3 is by no means the only great model out there, and I think it’s helpful to get good grounding in the bones of a prompt that work regardless of the model.

Yes, they exist! I did the research for you and put together a both a prompt anatomy and also a detailed explanation of why and how that prompt structure actually works (down to the level of token architecture). I’ve also included examples from recent AI papers that show how each prompt section measures up and drives value for the prompt as a whole.

My goal isn’t theoretical. I’m trying to help you understand why and how prompting works down to the token-level so that you feel more fluent when you’re wrestling with your own models—and so you feel more able to adapt prompts I share here into whatever context you’re operating in. Prompting is as much art as science, but you’ll see here that there’s plenty of science to dig into!

Read on for a prompting guide you can pin to the wall by the computer and use every day (I even made a table for that that’s easy to print lol).