Prompt writing: a 4-part formula that fixes bad answers

To write better prompts, give the AI four things in a clear order: a role (who it should act as), context (the facts about your situation), the task (what you want done), and the format (how the answer should be shaped). Most weak answers trace back to a prompt that was missing two or three of those pieces. Add them back and the same model hands you something you can use.

Most people type a prompt the way they'd text a friend who already knows the backstory. The AI has no backstory. It fills the gaps with the most average guess it can make, and average is exactly what you don't want.

Why do my prompts get generic answers?

Because a short prompt is an under-specified request, and the model answers the request you sent, not the one in your head. Ask "write me a welcome email" and you'll get a welcome email for nobody, in no voice, about no product. It's technically correct and completely useless.

The fix is not a longer prompt. It's a more complete one. Four ingredients cover almost everything, and once you know their names, you'll notice which one is missing every time an answer comes back flat. If you want the deeper reason this works, it comes down to how these models predict text, which I break down in how large language models work.

What are the four parts of a good prompt?

Here's the formula I teach every owner who tells me AI "doesn't get it." Say the parts out loud in this order and your hit rate jumps.

Role points the model at the right corner of what it knows. Context stops it from guessing. Task keeps it from wandering. Format saves you the second round of "no, shorter, no bullet points." Skip any one and you feel the gap in the answer.

What does a bad prompt look like fixed?

Let me take one weak prompt and rebuild it three times, adding parts as we go. Watch the answer get more usable at each step.

Start with the version most people type:

Write a social post about my new coaching program.

That gives you a post for a generic coach selling a generic program. Now add a role and a task that means something:

You are a social media copywriter with a warm, plain-spoken human voice.
Write one Instagram caption announcing my new coaching program.

Better voice, still hollow, because the model knows nothing about the program. Now add context and format, and you have a prompt that returns something you'd post:

You are a social media copywriter with a warm, plain-spoken human voice.
Here is my context: I coach first-time managers who just got promoted and feel
in over their heads. My program is an 8-week group cohort, starts [date], costs
[price], and the promise is "lead your team without working nights and weekends."
My audience is on Instagram, mostly women 28 to 40, skeptical of hype.
Task: write one Instagram caption announcing the program.
Format: 4 short lines plus a one-line call to action, no hashtags, warm but not
salesy, and do not invent any results or testimonials I did not give you.

Same model, same thirty seconds of typing. The third answer is a draft you edit. The first is a draft you delete.

How much context is enough?

Enough that a smart freelancer could do the job without asking you a follow-up question. That's the bar. If you'd have to answer "who's this for?" or "how long?" before a human could start, the model needs that answer too.

A designer I'll call Marcus timed his prompt habit for a week. He was averaging four back-and-forth messages to get one usable answer, because he started thin and patched as he went. We rebuilt his top five requests using the R-C-T-F formula, front-loading the context. His round-trips dropped to about one and a half.

He wasn't a better prompt writer overnight. He just stopped making the model guess. Every guess it gets wrong is a round-trip you pay for in time.

A vague prompt isn't faster. You just pay the cost in follow-ups instead of up front.

The other reason to front-load context: shorter total conversations. Every patch message re-sends the whole thread, which burns through your usage faster than one complete prompt would. That tradeoff is worth understanding, and I cover it in why token efficiency saves you money.

Once you can write one strong prompt, the next move is to stop rewriting it from scratch every time. That's the whole idea behind reusable prompt templates.

If you'd rather start from prompts that already have all four parts built in, the WorkSmart prompt packs are $29 one time and give you 25 fill-in-the-bracket prompts across leadership, growth, content, and productivity. You swap in your specifics and go.

Do this next

Take the last prompt you typed that gave you a flat answer and rewrite it with all four parts, out loud, in order: role, context, task, format. Run it again and compare. When you want the version where the hard part is already done, the WorkSmart prompt packs hand you 25 tested prompts you only have to fill in.