There is a whole industry selling prompt templates. You can safely ignore almost all of it. The thing that actually separates a useless answer from a genuinely good one is not a magic phrase. It’s context — and you already have all of it, sitting in your head, unstated.
Why the answers feel generic
When you ask a vague question, the assistant does the only sensible thing available to it: produces the most average possible answer. Average is what you asked for. It doesn’t know your situation, your reader, your history, or your constraints, so it writes for nobody in particular.
That is not a limitation of the technology. It’s a straightforward consequence of the request.
The four things worth saying
You don’t need a formula. But nearly every genuinely good request contains some of these, and nearly every disappointing one contains none.
- What it’s for. The actual purpose, not the format. “A letter” is a format. “A letter that gets my deposit back without burning the relationship” is a purpose.
- Who it’s for. A stranger, a friend, your accountant, someone annoyed with you. Audience changes everything.
- What you already know. Constraints, history, what you’ve tried, what definitely won’t work.
- What wrong looks like. Often the most useful of the four, and the one nobody thinks to include. “Don’t make it chirpy.” “Don’t apologise.” “Not a listicle.”
The same request, twice
“Help me write a bio.”
You will get something serviceable, forgettable, and true of roughly four million people.
“Help me write a 60-word bio for a gallery show. I’m a painter, mostly oceans, and I’ve also spent thirty years building software — the two are related and I want that to come through without sounding like a tech person who paints on weekends. Understated. No adjectives like ‘passionate’ or ‘award-winning’.”
Same tool, ninety seconds of typing, entirely different result. You didn’t learn a technique. You just said the things you already knew.
When it’s hard to explain
Sometimes you can’t articulate what you want — you’ll know it when you see it. Two things work well here.
Show, don’t describe. Paste in an example of something you like and say “this sort of thing, but for my situation.” Examples carry more information than adjectives ever will.
Make it ask you. Try: “Before you write anything, ask me five questions that would help you do this well.” This is the single most underused move available, and it frequently surfaces things you hadn’t considered.
Treat the first answer as a draft
The first output is a starting position. React to it the way you’d react to a first draft from a capable person who doesn’t yet know your preferences: say what’s wrong, keep what works, ask again.
Nearly everyone who finds this technology disappointing stopped after round one.
The short version
If you remember nothing else: say what it’s for, who it’s for, and what would be wrong. That’s the whole guide. Everything above is elaboration.
Common questions
Do I need to learn prompt engineering?
No. The paid prompt-template industry is mostly theatre. Explaining what you want the way you would explain it to a competent new assistant — what it is for, who it is for, and what you have already tried — outperforms nearly every template you can buy.
Why are my results so generic?
Almost always because the request contained no context. The assistant defaults to the most average possible answer when it does not know the situation, the audience, or the constraints. Adding two sentences of context typically changes the result more than switching tools would.