How to write prompts that get useful AI answers

To write a good prompt, tell the AI five things: who the answer is for, what you want it to do, the context it needs, the format you expect, and any limits such as length or tone. Most weak answers come from missing context, not from a weak model. The method below works in ChatGPT, Claude, Gemini and other assistants.

The five parts of a good prompt

You rarely need all five for a quick question, but for anything you will use, the more of them you include, the less editing you will do.

Before and after examples

Weak: “Write a LinkedIn post about remote work.” Better: “Write a LinkedIn post for HR managers about one lesson from moving our 40-person company to remote work: weekly written updates replaced most status meetings. First person, 120 to 150 words, end with a question to readers, no hashtags.”

Weak: “Explain inflation.” Better: “Explain inflation to a 15-year-old using an example about phone prices. Then give a one-sentence textbook definition.”

The improved prompts add audience, specific content, format and limits. That is the whole technique.

Give examples of what good looks like

Showing beats describing. Paste a paragraph you like and say “match this tone”, or give two examples of the output format before asking for more. This is called few-shot prompting and is one of the most reliable ways to get consistent results.

Break complex tasks into steps

For a big task, ask for one stage at a time: first an outline, then the first section, then a review. Or ask the model to plan before it writes: “First list the steps you will take, then do them.” Reasoning models do this internally, but explicit steps still help you check the work as it goes.

Let the AI ask you questions

When you are unsure what context matters, end with: “Before you answer, ask me up to five questions that would improve your response.” The model often asks about exactly the details you forgot.

Ask for evidence and uncertainty

For factual tasks, add “Quote the source for each claim” or “Say if you are unsure.” That makes AI hallucinations easier to spot. When you supply a document, tell it to use only that document.

Iterate instead of starting over

Treat the first answer as a draft. Tell the model what to change: “shorter”, “more concrete examples”, “less formal”, “the second point is wrong because…”. Specific feedback works better than rewriting the whole prompt.

Common prompt mistakes

1. Asking several unrelated things in one message.
2. Leaving out the audience, so the answer is pitched at the wrong level.
3. Not saying the format, then reformatting by hand.
4. Pasting huge documents when only one section matters, which wastes tokens.
5. Accepting the first answer for important work without checking it.

Save prompts you reuse

If you write the same kind of prompt often, save it as a template, add it to ChatGPT’s custom instructions, or build a custom GPT or Claude Project around it. For ready-made templates, see our list of ChatGPT prompts, and for pictures, our guide to AI image prompts.