Stop Re-Explaining Yourself to Your AI — Write It a Rule

I kept having the same argument with my AI assistant. Its answers were technically correct and genuinely exhausting to read — dense, jargon-heavy, no clear takeaway. So every few messages I’d add “keep it simple” or “just give me the point,” get one good reply, and then watch it drift right back to a wall of text in the next conversation.

At some point I noticed the pattern in my own behavior: I was giving the same feedback over and over. And repeated feedback isn’t a preference you express — it’s a rule you haven’t written down yet.

Encode it once

So I sat down and wrote the assistant an actual rule for how it should talk to me. Not what to do — how to say it. Plain language over jargon. A warm tone. Breathing room instead of dense blocks. Examples and analogies where they help something land. Visuals — a table, a few steps, a small diagram — but only when they genuinely cut the effort of reading, never as decoration.

The important refinement was tailoring by type of response. A status update should be terse. Something I’m trying to learn should be richer, with the analogy and the example. One rule, but context-aware, so it isn’t uniformly verbose or uniformly clipped.

Since then I don’t re-explain. The preference lives in one place and applies every session. The recurring friction just… stopped.

A wrinkle worth knowing

If you go to do this, know that these tools tend to store your rules in two different places, and they behave differently. Some live in an editable rule file the assistant can read and even revise for you. Others are locked in a settings block inside the app’s own storage — which means the assistant can’t edit those; you have to paste them in yourself. Worth knowing before you wonder why your carefully-worded instruction didn’t take.

”Short” is not “clear”

The subtlest thing I learned came from a collision between my new rule and an older one I’d written that just said be as short as possible. Those two rules fought. The style rule wanted room to breathe and an occasional example; the old rule wanted the fewest possible words.

The resolution reframed what I actually wanted. The real target was never fewest words — it was lowest effort to understand. Sometimes that means fewer words. But sometimes a one-line example or a small table makes something click instantly that a terse sentence would have made me re-read three times. Optimizing for word count can actively make an answer worse. “Short” and “clear” point in the same direction often enough to feel identical — and then, right when it matters, they don’t.

So: if you keep telling your AI the same thing, stop telling it and write it a rule. And when you do, aim the rule at clarity, not brevity. They’re not the same goal, and only one of them is the one you actually care about.