Prompt Engineering Basics That Actually Improve Answers
A prompt is simply what you say to a language model. Prompt engineering is the craft of saying it well. The phrase sounds technical, but the underlying skill is familiar to anyone who has briefed a new colleague: the clearer the brief, the better the work. A handful of habits account for most of the improvement.
The model does exactly what you ask, and nothing you did not
The biggest source of disappointing answers is a vague request. "Write something about our hotel" gives the model nothing to aim at, so it produces generic filler. The model is not being lazy. It is filling in every unspecified choice with the most average option, because average is the safest guess.
Prompting well means making those choices yourself, up front.
The four parts of a good prompt
Most strong prompts contain four ingredients. You do not need all four every time, but each one closes a gap the model would otherwise fill with a guess.
A role. Tell the model who it is for this task. "You are a concierge at a boutique hotel" sets a tone, a vocabulary, and a point of view far more efficiently than describing all three.
A task. State what you want in plain terms, including what the output is for. "Write a two paragraph welcome note for guests arriving by train" beats "write a welcome note".
An example. If you have one, show a sample of what good looks like. Models imitate patterns extremely well, and one example often does more than a paragraph of description. Showing a few examples is sometimes called few shot prompting.
A format. Say how the answer should be shaped: a list, a table, a fixed length, a particular structure. Models follow formatting instructions reliably, and a clear format makes the output usable without editing.
Give it the facts
A model cannot know your check in time, your cancellation window, or your chef's name. If the answer depends on facts, put the facts in the prompt. This sounds obvious and is the step most often skipped. Many "hallucinations" are simply a model asked to be specific without being given anything specific to work from. Supplying the relevant facts, or connecting the model to a source of them, is the single most effective way to get accurate answers.
Ask it to think, and to admit limits
Two small instructions improve reliability. First, for anything involving reasoning, ask the model to work through the problem step by step before giving its conclusion. Laying out the steps tends to produce better conclusions. Second, give it permission to say it does not know. A line like "if the information is not provided, say so rather than guessing" cuts invented details sharply, because it makes "I do not know" an acceptable output.
Iterate instead of perfecting
Prompting is cheap to repeat. The right workflow is to write a reasonable prompt, look at the answer, notice what went wrong, and adjust one thing. Wrong tone, adjust the role. Wrong shape, adjust the format. Wrong facts, supply the facts. A few rounds of this beats hours spent crafting the perfect first attempt.
Where prompting stops
Prompting controls behavior and shape. It is the right tool for tone, format, and reasoning style. It is the wrong tool for teaching a model a large body of knowledge, because everything in a prompt has to be supplied every time and there is only so much room. When the problem is knowledge rather than behavior, the answer is retrieval, giving the model a library to draw from, not a longer prompt.
The takeaway
Treat the model like a capable new hire who knows nothing about your business. Tell it who it is, what you need, what good looks like, and what shape to deliver it in. Hand it the facts. Let it say when it is unsure. Most of prompt engineering is just that: being clear.