Prompting.

Why you still have to decide every time.

By Jan-Willem Bekkers · July 27, 2026 · 4 min read

None of what this series has covered is hard to understand. Context, Role, Norm, the trade-off underneath all three, you've read the reasoning, you could probably explain most of it to a colleague by now. The gap was never understanding it. It's doing it, on purpose, for every single prompt, on a day when you've already got eleven other tabs open.

A normal Tuesday

Say you run a small online shop. By eleven in the morning you've typed four separate prompts into an AI, and each one needed something different from you, not because you were inconsistent, but because the tasks themselves were:

Four prompts, four different answers to which two or three letters actually carried the weight. Nothing about any of those calls was wrong, you made each one correctly, probably without consciously framing it as a decision. But you made it four separate times before lunch, and it's only eleven.

The part that doesn't scale is the repetition, not the reasoning

Knowing the rules isn't enough. argued the hard part of prompting was never memorizing Context, Objective, Role, Drive, Output, Norm, it was deciding how much weight each one deserved for the task in front of you. That's true for one prompt. It's still true for the fortieth prompt of the week, except now it's the same judgment call, made fresh, again, at three in the afternoon, when you're behind on the actual work the prompt was supposed to help with in the first place.

That's not a competence problem. Why you get bad AI answers. opened this series on the idea that the model isn't the problem and neither are you, and that holds here too. Knowing what a good prompt needs doesn't fail because anyone stops understanding it. It fails the way any manual judgment call fails under repetition, not from a lack of skill, but because doing something well once and doing it well forty times this week draw on two completely different things. A systematic study across ten different software-engineering tasks found the same pattern in a narrower, more technical setting: no single prompting technique won across the board, which one worked best depended on whether the task needed step-by-step logic or just relevant examples. The same task-dependency shows up in everyday prompting, just without the benchmark: getting a good result is rarely about a fixed formula, it's about what this specific task needs stated, and that judgment call resets with every single prompt.

What actually changes

That supplier message above is a good example of a field genuinely not applying: a one-line stock question doesn't have a real Role or Norm behind it. CORDON's own minimum, four of the six fields, reflects exactly that. It doesn't manufacture an answer for a field with nothing real to say, it builds the finished prompt from whichever ones genuinely carry the task, whether that's two of them or all six.

This is the actual gap CORDON closes, and it's a narrower, more honest claim than “CORDON writes better prompts than you.” You already know, after this series, what each of those four Tuesday-morning prompts needed. Try CORDON, and what changes is that acting on that knowledge takes a few seconds instead of a minute of quietly deciding, every single time, on the day you have the least room left to do it by hand.

Build your own custom AI prompt in seconds.

Try CORDON

3 free improvements. No account required.

Why you still have to decide every time · CORDON