How to write clear instructions for AI

A prompt is a work order, and the clarity you would give a competent new employee is exactly what the machine needs.

The market sells prompt folklore: magic words said to unlock hidden modes, perfect phrases copied and pasted as if they worked everywhere, and prompt collections traded like charms. An entire industry of shortcuts has grown around a skill that is older than any of the tools involved.

The honest version is simpler. A prompt is a work order, and the machine resembles a competent new employee on the first morning: no context, no memory of your intentions, no ability to guess what a short sentence left out. Once you accept that, the hunt for the perfect phrase ends and the ordinary craft of giving instructions begins.

The practice that transfers starts with two statements: what the task is, and what done looks like. Then give the material the answer must stand on, whether that is your data, your draft, or the documents the work should reflect. A machine asked to argue without material will invent some, and invention is rarely what you wanted.

Next, state the constraints: the length, the audience, the language, and what must never appear in the output. Show one example of the shape you want, because a single honest example replaces rounds of correction. None of this is cleverness; it is the same briefing a good manager writes without thinking.

Then judge the output against the instructions before iterating. If the result is wrong, the instruction usually was: missing context, a vague verb, two goals folded into one request. Reading your own instructions the way a stranger would is the entire skill, and it improves faster than any template.

We write such instructions daily inside our own operation, where the systems that serve our clients run on briefings we draft, review, and correct like any colleague's work. We do not sell spells. We tell you plainly that clear work orders are the skill, and that the skill stays useful no matter which tools come next.