How to write prompts that get better AI answers

A good prompt is not a spell

Prompting is not about secret phrases. It is about giving the model a clear task, useful context and criteria for what a good answer should look like.

The biggest beginner mistake is asking for an output without explaining who it is for, why it matters and how to judge whether it is good.

A simple prompt structure

A useful prompt often includes five parts:

1. Goal: what you want to achieve.
2. Context: audience, situation and constraints.
3. Input material: text, data, examples or background.
4. Output format: list, table, plan, draft, code or analysis.
5. Quality criteria: tone, length, level of detail and what to avoid.

Not every prompt needs all five. But when answers are weak, one of these pieces is usually missing.

A weak prompt

Write a post about AI.

This is too broad. The model does not know the audience, goal, tone, length or angle.

A better version

Write a short LinkedIn post for small business owners. Topic: how to use AI to reply to customer inquiries faster. Tone: practical, no hype. Length: 1,200-1,500 characters. Include 3 concrete examples and end with a question for readers.

This prompt gives the model a role, audience, topic, tone, length, structure and desired outcome.

How to improve answers

The best results usually come through iteration. Do not expect the first answer to be perfect. Guide the model like you would guide a teammate:

• “Make it 30% shorter.”
• “Add more concrete examples.”
• “Remove the marketing tone.”
• “Make it more technical.”
• “List assumptions you are unsure about.”
• “Give me three versions and recommend the best one.”

AI models respond well to specific feedback. The clearer you are about what is wrong, the faster you get something useful.

Prompt for text analysis

Analyze the text below.
 
Goal: improve clarity and credibility.
Audience: non-technical readers interested in AI.
 
Return:
1. Biggest problems with the text.
2. Parts that sound too generic.
3. A revised version.
4. Facts that should be verified.
 
Text:
...

Prompt for comparing tools

Compare three AI tools: [tool A], [tool B], [tool C].
 
Evaluate them by:
- use case,
- ease of use,
- limitations,
- cost,
- data risk,
- best user type.
 
Return the answer as a table and add a short recommendation at the end.

What to watch out for

A prompt does not replace thinking. A model can sound confident and still be wrong. For important work, ask it to list assumptions, risks and points that require verification.

Also remember that not every problem is solved with a longer prompt. Sometimes it is better to split the task into steps: analysis first, plan second, draft third, editing last.

The simplest rule

If a prompt is unclear to a person, it will usually be unclear to a model. Start with the goal, add context, define the output format and improve the result in follow-up steps.

Related concepts: prompt engineering, large language model, AI hallucination.