“Prompt engineering” sounds like a specialized skill, but it is mostly common sense once you know how a language model works. Since a model just continues your text in the most plausible way, the way you ask shapes what you get. A vague prompt gives a vague answer. A clear, well-set-up prompt gives a sharp one. Here are the techniques that actually move the needle.
Be specific about what you want #
The most common mistake is being too vague and expecting the model to read your mind. It cannot. It fills ambiguity with the most generic plausible answer.
Instead of “write about dumplings,” try “write a 200-word introduction to a pork and chive dumpling recipe, aimed at a beginner cook, in a friendly but not cutesy tone.” Every detail you add (length, audience, tone, angle) removes a dimension the model would otherwise guess on.
Give it context and a role #
The model does better when it knows the situation. Two easy moves:
- State the context and the goal. “I am writing docs for non-technical users” tells the model how to pitch everything that follows.
- Give it a role. “You are an experienced line cook explaining this to a home cook” primes it to draw on the right patterns. This is not magic, it is steering the model toward the kind of text you want it to produce.
Show an example or two #
This one is underused and powerful. If you want output in a particular style or shape, show the model an example of it. Give it one or two samples of the format you want, then ask it to do the same for your new input. Showing beats telling: a single concrete example often does more than a paragraph of description, because the model is very good at matching a pattern you put in front of it.
Ask for the format you need #
If you want a table, ask for a table. If you want five bullet points, say five bullet points. If you want just the answer with no preamble, say “respond with only the answer, no introduction.” Models love to add throat-clearing (“Great question! Here is…”) and framing you did not ask for. Telling it the exact output shape you want cuts that out and gives you something you can use directly.
Break big tasks into steps #
For anything complex, do not ask for the whole thing in one shot. Either break it into a sequence of prompts, or tell the model to work through it in steps: “First outline the structure, then we will write each section.” Walking a hard problem step by step tends to produce better results than demanding the finished product immediately, because each step gives the model something concrete to build on.
Give the reason behind the request #
A quietly effective habit: tell the model why you want something, not just what. “Summarize this for a busy executive who needs to decide whether to fund the project” produces a very different, and more useful, summary than “summarize this.” When the model understands the intent, it makes better choices about what to include and how to frame it, instead of guessing at your purpose.
Say what to avoid, sparingly #
Telling the model what not to do can help (“do not use jargon,” “avoid the word ‘delve’”), but use it in moderation. A wall of negative instructions is less effective than a couple of positive examples of what you do want. When you can, show the target rather than listing everything to steer clear of.
Iterate #
The biggest mindset shift is to treat the first answer as a draft, not a verdict. If it is not right, do not start over from scratch, adjust. “Good, but make it shorter and less formal.” “Closer, but you missed the part about cost.” Prompting is a conversation, and steering toward what you want across a couple of turns is normal and expected, not a sign you did it wrong.
The whole thing in one line #
Good prompting is just clear communication with something that takes you literally and cannot read your mind. Be specific, give context, show examples, ask for the format, explain the intent, and refine. Do that and you will get dramatically more out of any model, and you will not need any tricks beyond understanding how the model works in the first place.