How to Write Better AI Prompts: A Simple Framework

If you want to know how to write better AI prompts, the secret is not fancy phrasing. It is giving the tool the same information a human helper would need: what you want, who it is for, what to use and how it should look. Most weak answers come from vague requests rather than weak technology. This guide offers a simple five-part framework you can remember, examples for everyday tasks, a way to improve results through short follow-ups and a habit of saving prompts that work.
The five parts of a good prompt
A strong prompt usually contains five parts. You can remember them as goal, audience, context, format and limits. State the goal as one clear action, such as summarize, compare, draft or explain. Name the audience, for example a busy manager or a ten-year-old. Add the context the tool needs, including facts, source text and background. Specify the format, such as bullet points, a table or three short paragraphs. Finally, set limits: length, tone, topics to avoid and what to do when information is missing. For a related overview, see prompt engineering basics.
Examples for everyday tasks
- Email: write a polite reply declining a meeting, offer two alternative times and keep it under eighty words.
- Summary: summarize the text below for a manager in five bullet points and list any decisions needed.
- Planning: create a one-week study plan for a beginner learning basic statistics, with thirty minutes per day.
- Comparison: compare these two phone plans in a table covering price, data and contract length.
More templates for work messages are in prompt templates for business emails.
Show an example
Models are very good at copying patterns, so a sample often beats a long description. Paste a short example of the style or format you want, and say, write in the same style but about this new topic. You can also give a pair of examples, one good and one poor, to explain what to avoid. Make sure the example does not contain private information, and remove any facts you do not want repeated. Examples are especially helpful for tone, such as friendly but professional, and for structured outputs such as product descriptions.
Iterate instead of restarting
Treat the first answer as a draft. Reply with specific edits: make it shorter, use simpler words, add a concrete example in the second paragraph, remove the jargon. Ask the tool to explain its reasoning or to list assumptions it made. If one section is good, say keep this and rewrite the rest. Each small instruction builds on the previous context, which is faster than rewriting the whole prompt. When a conversation becomes confusing, begin a new one and paste your best instructions.
Save prompts that work
Keep a notes file with prompts that produced good results. Add a title, the date, which tool you used and a short description of when it works well. Replace changeable details with brackets so you can reuse the prompt quickly. Share the best ones with your team, and update them as tools improve. Over time you build a toolkit that makes quality more consistent. Marketers can start with prompts for better marketing copy.
Mistakes that weaken prompts
Avoid piling several unrelated tasks into one message, because quality drops. Do not assume the tool knows your situation. Avoid contradictory instructions such as be brief but cover everything. Never include confidential data you would not share publicly. Finally, do not skip checking: even a well-written prompt can produce errors, so read how to check AI answers. Sometimes a plain question is enough, which we discuss in prompt engineering vs plain questions.
A before and after example
Before: help me with a presentation. After: act as a presentation coach and help me outline a ten-minute talk for new employees about our safety rules. The audience is not technical, so use plain words, three main points and one short story for each. Format the outline as a numbered list with a one-line speaker note per slide. Limit it to eight slides and ask me two questions if anything is unclear. The second version gives the goal, audience, context, format and limits, so the first reply is usable and the follow-up edits are small.
Quick checklist before you press send
- Is the goal one clear action?
- Did I name the audience?
- Did I include the facts or text the AI needs?
- Did I choose a format and a length?
Frequently asked questions
How long should a prompt be?
Long enough to include the five parts. Many good prompts are three to six sentences.
Should I say please or use polite language?
It does not matter much for results, so use whatever tone feels natural.
Why do I get different answers each time?
Models include some randomness. Clear instructions and examples make results more consistent.
Can the AI write the prompt for me?
Yes. Ask it to turn your rough idea into a clear prompt, then review and edit it.


