Most prompt engineering advice is either obvious or superstition. You do not need a 2,000-word mega-prompt or a secret phrase that unlocks the model true power. You need to say what you want the way you would say it to a sharp, literal-minded colleague who has no context and will take you at your word.
The four things that actually help
1. State the task, not the topic
"Write about our new feature" gets you mush. "Write a 120-word changelog entry for a feature that exports reports as PDF, aimed at existing customers, plain and slightly enthusiastic" gets you something you can use. The model is not guessing your intent for fun. It is guessing because you did not tell it.
2. Give it the context it cannot have
It does not know your product, your audience, or last week decision. Paste the relevant bit. A smaller model with the right paragraph of context beats a bigger model without it, almost every time. This is the single highest-return habit, and it is boring, which is why people skip it.
3. Show one example
One good example of the output you want is worth three paragraphs of description. Want a certain tone? Paste a sentence in that tone. Want a specific format? Show the format once. Models are excellent mimics and mediocre mind-readers.
4. Say how long and in what shape
Length, format, audience. "Three bullet points, one line each." "A short paragraph a non-technical manager would understand." If you do not specify, you get the model default, which is usually too long and vaguely corporate.
Things that do not work as well as people claim
- Bribing and threatening. "I will tip you $200" and "my job depends on this" made a measurable difference on older models. On current ones the effect is mostly gone. Save your breath.
- Role-play stacking. "You are a world-class expert with 30 years of experience" rarely beats describing the task well. One line of role setting is plenty.
- Magic words. There is no incantation. If a prompt only works with one specific phrase, it is fragile and will break on the next model version.
The habit that matters most
Iterate out loud. Your first prompt is a draft. Read the output, tell the model exactly what is wrong ("too formal, cut the intro, keep the second point"), and let it fix it. People treat prompting like a one-shot spell when it is a conversation. The best result usually arrives on the third message, not the first.
A prompt is not a magic phrase. It is a clear brief. Write it like you respect the reader time, and the machine will too.
One last thing, since we are being honest: if you cannot explain what you want in plain language, the model cannot either. Half of bad output is a thinking problem wearing a prompting costume. Fix the brief in your head first. The rest is easy.