Quick Takeaway
A good AI prompt tells the model who to be, what you want, why, for whom, and in what format. Give it a role, a clear goal, relevant context, the audience, constraints such as length and tone, and the output format. Specific prompts get specific answers; vague prompts get generic ones.
Key takeaways
- Strong prompts have six parts: role, goal, context, audience, constraints and output format.
- The more context you give, the less the AI has to guess, and the better the answer.
- Ask for a specific format (table, bullet list, JSON, email) to get output you can use straight away.
- If an answer is weak, fix the prompt: add missing context or an example rather than just asking again.
- Different models respond to slightly different structures; Claude, for example, works well with clearly separated sections.
Most disappointing AI answers come from short, vague prompts. “Write a blog post about CRM” gives the model almost nothing to work with, so it fills the gaps with the most generic answer possible. Add a few sentences of direction and the same model produces something you can actually use.
This guide shows you what to put in a prompt, with examples you can copy. If you want to skip the manual work, our free AI prompt generator builds a structured prompt from a one-line idea, for ChatGPT, Claude, Gemini, AI agents and image tools.
- AI prompt
- An AI prompt is the instruction you give a language model: the task, plus the context, constraints and format it needs to produce a useful answer.
The 6 Parts of a Good Prompt
| Part | What it does | Example |
|---|---|---|
| Role | Sets the expertise and perspective | “You are an experienced B2B email marketer.” |
| Goal | States exactly what you want | “Write a 3-email welcome sequence.” |
| Context | Gives the facts only you know | “We sell accounting software to small agencies.” |
| Audience | Says who the output is for | “Owners of 5–20 person agencies, new to the product.” |
| Constraints | Limits length, tone and content | “Under 120 words each. Friendly, no jargon.” |
| Output format | Makes the answer usable | “Give each email a subject line and body.” |
You won’t need all six every time. A quick question can be one sentence. But for anything you plan to publish, send or build on, include as many as apply.
A Before and After Example
Here is the same request written two ways.
Write a LinkedIn post about automation.You are a B2B content writer who explains technology in plain language.
Write a LinkedIn post for owners of small service businesses (5–30 staff) who spend hours on manual follow-ups.
Goal: show how automating lead follow-up saves time, and invite readers to comment with their biggest admin task.
Context: we build automations with n8n and AI. Do not mention prices.
Constraints: 150–200 words, short paragraphs, one concrete example, no hashtags in the body, max 3 hashtags at the end, no buzzwords such as "game-changer".
Output: the post text only.The second prompt tells the model who it is writing for, what the post must achieve, what to avoid and how long it should be. The answer will be far closer to something you would publish.
How to Prompt ChatGPT, Claude and Gemini
The six parts work everywhere, but each model has habits worth knowing. The vendors’ own guides, linked below, are the best reference.
| Model | What tends to work well |
|---|---|
| ChatGPT (OpenAI) | Clear instructions, headings or delimiters between sections, and asking it to take a step-by-step approach for complex tasks. |
| Claude (Anthropic) | Clearly separated sections (Anthropic recommends XML-style tags such as <context> and <instructions>), plus examples of the output you want. |
| Gemini (Google) | Specific instructions with examples, and explicit references to any files, images or documents you attach. |
| Image models | A single descriptive paragraph: subject, style, composition, lighting, colors, plus what to avoid. |
Show, don’t just tell
One good example of the output you want (“Here is a post I liked: …”) often improves results more than extra instructions. Label it clearly so the model imitates the style, not the content.
Prompt Templates You Can Copy
You are an expert [field] analyst.
Task: [what you want analyzed or answered].
Context: [background, data or constraints].
Audience: [who will read this].
Structure the answer as:
1. A 2-sentence summary
2. Key findings as bullet points
3. Risks or unknowns
4. Recommended next steps
If information is missing, say what you would need instead of guessing.You are an experienced [type] writer.
Write a [format] about [topic] for [audience].
Goal: [what the reader should think or do afterwards].
Tone: [e.g. friendly, expert, direct]. Length: [words].
Include: [must-have points]. Avoid: [phrases, claims, topics].
Return only the final text.Improve the text below for clarity and flow.
Keep every fact, number and name unchanged. Do not add new claims.
Shorten long sentences and replace jargon with plain words.
Then list the three biggest changes you made.
TEXT:
"""
[paste your text]
"""How to Fix a Prompt That Isn’t Working
Step 1: Find what is wrong with the answer
Too generic? Wrong tone? Too long? Missing facts? Name the problem before you rewrite anything.
Step 2: Add the missing context
Generic answers usually mean the model lacks information. Add who it is for, what you already know and what you will use it for.
Step 3: Constrain the format
Ask for a table, a numbered list or a word limit. Structure removes rambling.
Step 4: Give an example
Paste a short example of a good output and say what you like about it.
Step 5: Ask it to check itself
End with “Before answering, check that you have met every requirement above.”
Common Prompt Mistakes
- One-line prompts for complex tasks. Complex work needs context.
- No audience. “Explain X” to whom? A beginner and an expert need different answers.
- No format. You get paragraphs when you needed a table.
- Asking for facts without sources. Ask the model to say when it is unsure, and verify important claims.
- Pasting private data carelessly. Check your company’s rules before sharing customer or confidential information with any AI tool.
Prompts for AI Agents and Automation
When a prompt runs inside an automation or an AI agent, nobody is there to correct it, so it needs to be more precise: define the allowed actions, the exact output format (often JSON), what to do when data is missing, and when to hand over to a person. Our guide to AI chatbots versus AI agents explains the difference, and our AI automation services build these systems end to end.
Once your content is drafted, you can check whether it sounds natural with our guide on making AI writing sound human. For ready-made examples, see 30 ChatGPT prompts for small business.
What is the best way to write an AI prompt?
Give the model a role, a clear goal, relevant context, the audience, constraints (length, tone, what to avoid) and the output format. The more specific the prompt, the more useful the answer.
How long should a prompt be?
As long as it needs to be. A simple question can be one line; a piece of writing or analysis often works best with 5–15 lines of context and requirements.
Do the same prompts work for ChatGPT, Claude and Gemini?
Mostly, yes. The core parts are the same, but formatting helps: Claude works well with clearly separated sections, and Gemini benefits from explicit references to attached files.
Why does the AI give generic answers?
Usually because the prompt lacks context. Add who the output is for, what you already know, and an example of what good looks like.
Is there a free AI prompt generator?
Yes. Our AI prompt generator is free, needs no sign-up, and builds structured prompts for 25 task types and 8 AI destinations. It can also improve a prompt you already have.
Should I trust facts in AI answers?
Verify anything important. Ask the model to flag uncertainty and to cite sources, and check those sources yourself.