How to Write Better AI Prompts: A Complete Guide (With Examples)
The short version
Good prompts share five things: a clear role, a specific task, relevant context, a defined format, and an example or two. Vague in, vague out. The single biggest upgrade most people can make is simply telling the AI who it is, exactly what you want, and what good looks like.
What's inside
Most people who feel let down by AI tools don't have a bad tool — they have a vague prompt. The same model that produces generic, forgettable output for one person produces sharp, useful work for another, and the difference is almost entirely in how the request is written. The good news: prompting isn't a dark art. It's a handful of habits anyone can learn in an afternoon, and this guide gives you all of them, with real examples you can copy.
Why prompts matter so much
An AI model doesn't know what's in your head. It only has the words you give it, plus its general training. When you type "write me a marketing email," the model has to guess at everything you left out: who it's for, what you're selling, what tone you want, how long it should be. It fills those gaps with the most average, generic choice — which is exactly why the output feels bland. A strong prompt removes the guessing. The more of the right detail you provide, the less the model has to invent, and the closer the result lands to what you actually wanted.
The 5-part prompt framework
You don't need to memorise tricks. Almost every great prompt includes some combination of these five elements. Use as many as the task needs.
1. Role — tell it who to be
Asking the AI to adopt a role focuses its output. "You are an experienced email marketer" produces sharper marketing copy than no role at all, because it primes the model toward the right knowledge and tone.
2. Task — say exactly what you want
Be specific about the deliverable. Not "help with my email," but "write a 120-word promotional email announcing a 20% weekend sale." Specificity is the single highest-leverage habit in prompting.
3. Context — give it the background
Tell it what it needs to know: who the audience is, what you're selling, any constraints, the situation. Context is what turns generic output into your output.
4. Format — define the shape of the answer
Say how you want the result structured: a bulleted list, a table, three options, a specific length, a particular tone. If you don't specify, you get the model's default — which may not be what you need.
5. Examples — show what good looks like
One or two examples of the style or output you want will improve results more than almost anything else. "Match the tone of this example: [paste]" is remarkably effective.
Before & after: real examples
The fastest way to internalise this is to see weak and strong prompts side by side.
Example 1 — A marketing email
Write a marketing email for my sale.You are an experienced email marketer. Write a 120-word promotional email for a small independent coffee roaster announcing a 20% off weekend sale on all beans. Friendly, warm tone, no hype or exclamation marks. Include a clear subject line and one call-to-action button text.The strong version specifies role, length, business, offer, tone, and exact deliverables. The model no longer has to guess — and the output is usable on the first try.
Example 2 — Summarising a document
Summarise this.Summarise the document below for a busy manager who has two minutes. Give me five bullet points covering the key decisions and any action items, then one sentence on what it means for our team. Plain language, no jargon."Summarise this" gives you a summary of unknown length and focus. The strong prompt defines the reader, the length, the structure, and what to emphasise.
Example 3 — Brainstorming
Give me some business name ideas.Suggest 15 business name ideas for a mobile dog-grooming service aimed at busy urban professionals. I want names that feel friendly and trustworthy, are easy to say, and ideally have an available .com. Group them into "playful," "premium," and "simple," and avoid puns.Advanced techniques that genuinely help
Ask it to think step by step
For anything involving reasoning, math, or multi-step logic, adding "think through this step by step before giving your answer" measurably improves accuracy. It nudges the model to work the problem rather than blurt the first answer.
Iterate, don't restart
Treat it as a conversation. If the first draft is close, say what to change — "make it shorter and more formal," "lead with the benefit" — rather than starting over. Refining in steps gets you further than one perfect prompt.
Give it a way out
Tell the model what to do when it's unsure: "If you don't have enough information, ask me before guessing." This reduces confident-but-wrong answers, especially for factual tasks.
Assign constraints explicitly
"Use no buzzwords," "keep every sentence under 20 words," "only use information from the text I provide" — clear constraints shape output far more reliably than hoping the model infers them.
Common prompting mistakes
| Mistake | Fix |
|---|---|
| Being vague | Specify the deliverable, length, and audience |
| No context | Explain the situation and what you're trying to achieve |
| Asking for too much at once | Break big tasks into steps |
| Not specifying format | Say "as a table," "5 bullets," "200 words" |
| Accepting the first draft | Iterate — tell it what to change |
| Trusting facts blindly | Verify anything important; use a sourced tool for research |
That last point matters: even a perfectly prompted model can state something wrong with confidence. For research where accuracy counts, a source-backed tool like Perplexity lets you verify (see Perplexity vs ChatGPT).
Three templates you can copy
For writing: "You are a [role]. Write a [length] [type of content] for [audience] about [topic]. Tone: [tone]. Include [specific elements]. Avoid [what to avoid]."
For summarising: "Summarise the text below for [who]. Give me [format, e.g. 5 bullets] covering [what to focus on]. Keep it [length/style]."
For analysis: "Think step by step. Given [information], help me [goal]. Lay out your reasoning, then give a clear recommendation. If you need more information, ask first."
Which model you use matters less than how you prompt it — but quality does vary. If you care most about natural writing and following nuanced instructions, see our best AI writing tools guide and ChatGPT vs Claude comparison.
The bottom line
Better prompting is the highest-return skill in using AI, and it costs nothing to learn. Tell the model who to be, exactly what you want, the context it needs, the format of the answer, and an example of "good." Then iterate. Do that, and the same tools that frustrate most people will quietly become some of the most useful in your day.