A prompt with slots in it is a tool; one you retype is a habit

Retyping is where the variation comes from

You write more or less the same instruction each week, and it is never quite the same. One week you mention the audience, another you forget. Sometimes you ask for a word count, sometimes not. The output varies accordingly, and the variation gets attributed to the tool rather than to the twenty per cent of the instruction that changed.

A template removes that source of variation entirely. The fixed part stops being retyped, the variable part is explicit, and any difference in the output can be traced to the material rather than to whatever you happened to include today.

This is what converts a technique into something reusable. An instruction that exists only in your typing habits cannot be improved cumulatively, because each improvement lasts until the next time you write it out from memory slightly differently.

Name the blanks and say what fills them

A template is the fixed wording plus clearly marked slots — the audience, the source document, the maximum length, the deadline, the tone. Marking them distinctly, so a blank one is obvious at a glance, matters more than the notation you choose.

Each slot deserves a note about what belongs in it and what happens when it is empty. If no audience is given, assume an informed general reader. If no length is given, aim for three hundred words. Filling in a default explicitly is much safer than letting the gap be resolved silently on every run.

Keep the slot count small. A template with fifteen blanks is a form, and people fill forms in carelessly or abandon them. Three to five variables covers most recurring tasks, and anything that never actually changes should be written into the fixed part where it cannot be omitted.

A template is testable in a way an ad-hoc prompt is not

Because the wording is fixed, you can run it against a set of saved inputs and see what it does — a typical case, an unusually short one, one missing a field, one twice the normal length. That is a test suite, in the ordinary sense, and it takes an afternoon to assemble.

The payoff arrives when something needs fixing. A correction goes into one place, gets run against the saved inputs to confirm it did not break the previously working cases, and applies to everyone using it from that point. None of this is possible when the instruction lives in individual memory.

Keep a note of what changed and when. Templates accumulate clauses added in response to a single bad output, and without a record it becomes impossible to tell which clauses are load-bearing and which were a reaction to one unlucky run that nobody has revisited.

The real product is that someone else can use it

A colleague who has never thought about prompting can fill three labelled blanks and get the output your team expects. That is a much larger effect than any individual improvement in wording, because it moves a capability from one person to a group without training anybody.

Write templates to be read by that person. Plain wording, no accumulated superstition, a one-line description of what it produces and what it is not for. If a clause exists for a reason nobody remembers, either document the reason or remove it and see whether anything breaks.

The limits belong in the description too. Say what inputs this is not suitable for, and what a person must check before the output is used. A template that quietly implies the result is ready to send is worse than no template, because it transfers the work without transferring the judgement.

When a template is the wrong shape for the work

Exploratory work resists templating and should not be forced into it. Thinking through a problem, arguing with an idea, working out what you are trying to say — these are conversations rather than procedures, and a form would remove the exact flexibility that makes them valuable.

Tasks that recur but differ substantially each time are also a poor fit. If half the fixed wording has to be edited on each use, it is not a template, and maintaining the pretence is worse than writing the instruction freshly with a checklist of things to remember.

The honest test is frequency and sameness. If you have done a task the same way three or four times, write it down as a template. If you have done it three times in three different ways, the shape of the task is still unsettled, and freezing it now would fix the wrong version.