How to write a clear, specific AI prompt

Give the model the context it cannot guess

A model answers from what is in the conversation plus what it learned during training — it does not know who you are writing for, what you already tried, or why the task matters, unless you say so. "Rewrite this email" and "rewrite this email so a customer who is already frustrated reads it as an apology, not an excuse" are the same task with a different amount of context, and they get different answers. Context does not need to be long; it needs to include the one or two facts that would change how a person would do the task.

State the constraints instead of hoping they are implied

A length limit, a reading level, a tone, a thing to avoid mentioning, a deadline the answer has to respect — a model applies a constraint if you state it, and otherwise falls back to a generic default that may not fit. "Keep it under 150 words" and "avoid technical jargon" are both constraints a model can follow reliably once they are explicit; neither is something it infers correctly on its own with any consistency.

Name the output format you actually want

A bulleted list, a short paragraph, a table, a subject line plus a body — asking for the shape you want up front saves a follow-up message asking for a reformat. This matters more, not less, once the output has to be parsed by something other than a person reading it; for that case, see prompting for structured output, linked below, which is the deeper companion guide to this point.

Add an example when a description alone would be ambiguous

Some things are easier to show than describe — a house style, a tone, a specific format for a recurring task. One well-chosen example often resolves an ambiguity that several sentences of description would not. See few-shot prompting, linked below, for how to use more than one example deliberately, and when it is worth the extra length in the prompt.

Questions people ask

Does a longer prompt always get a better answer?

No — a longer prompt only helps if the extra length is context, a constraint, or an example the model would otherwise be missing. Padding a prompt with restated instructions or filler does not improve the answer and can bury the part that mattered.

Will a clear prompt stop a model from getting facts wrong?

No. A clear prompt makes it more likely the model understands what you are asking for, but it does not verify facts, and it can still state something false with full confidence — see why AI hallucinates, linked below, for what actually causes that and why prompting alone cannot fix it.

What is the single most useful thing to add to a vague prompt?

Usually context: the one or two facts about the audience, the goal, or the situation that a person would need to do the task well. A constraint or an example helps too, but they matter less if the model still does not know who the answer is for.