Practical guide · verified against the real thing
Writing better prompts in plain English (no incantations, no magic phrases)
In one line: You do not need secret wording — you need to give context, say what good looks like, and iterate. Here is the practical craft that reliably gets more out of any assistant.
Prompt "engineering" is often sold as a set of magic phrases. It is not. A model is a next-word predictor responding to the text you give it, so better prompts are not incantations — they are clearer communication. The same things that would help a capable human help a model: context, a clear ask, and feedback. None of this requires jargon.
The four moves that do most of the work
Give context — who you are, who the audience is, what it is for; the model cannot guess what you have not said. Be specific about the output — length, format, tone, what to include and leave out; "a 150-word email, warm but brief, no jargon" beats "write an email." Show an example when you can — one sample of the style or structure you want is worth a paragraph of description. And iterate — treat the first answer as a draft and refine, because a conversation narrows in on what you mean far faster than one perfect instruction.
Why this works (it is the mechanism)
Because the model predicts what should come next, the more precisely you describe the situation and the desired result, the more the plausible continuation is your desired result. Vague in, generic out; specific in, useful out. This is a direct consequence of how the models work — you are steering a predictor with context, not unlocking it with a code word.
What actually limits the result
No prompt fixes a model's blind spots: it will still not reliably know specific facts (why it hallucinates) and cannot verify its own claims, so prompt well and check the load-bearing details. Be careful what you put in the prompt too — the privacy rules in what not to paste apply to every prompt you write. And remember the tool underneath matters: which assistant you use is a separate question, worked through in AI assistants compared. Clear thinking in, useful work out — that is the whole craft.
Next