Prompt engineering is the skill of writing instructions that help a generative AI tool give you a useful answer. You do not need to code. You need to be clear about what you want, who it is for and what a good result looks like.
Below are seven before-and-after prompt examples you can copy and adapt, from writing an email to sorting messages with examples. Each one shows a weak prompt, a stronger one and why the stronger one works. The people, products and details are made up, so replace them with your own.
AdvertisementUse the free prompt builder after the examples to create your own prompt, then copy it into your favorite AI tool. AI answers can still be wrong, so always check them.
Prompt Engineering Examples
Quick jump: EmailSummaryStudy planResumeCodeProductExamples
Email to a Manager
Stronger prompt: You are a professional assistant. Write a short, polite email to my manager, Dana Lewis, asking to move our Friday meeting to Monday at 10 a.m. because I have a client call. Keep it under 100 words and end with a thank-you.
Why it works: It gives a role, the reader, the reason, the new time, a length limit and a tone.
Article Summary
Stronger prompt: Summarize the article below in five bullet points for a busy manager. Start each bullet with a verb, keep each under 20 words, and end with one sentence on what to do next. Article: paste the text here.
Why it works: It says how many points, who they are for, how long each should be and where the text begins.
Study Plan
Stronger prompt: You are a Spanish tutor. Create a four-week plan for a beginner who has 30 minutes a day. Use a table with columns for week, daily focus and a small goal. Include one speaking exercise each week.
Why it works: It sets a role, a time frame, the learner’s time, a table format and a required element.
Resume Bullet Points
Weak prompt: Improve my resume.
Stronger prompt: You are an experienced recruiter. Rewrite these three resume bullet points for a customer service role. Start each with a strong verb, add a number only if I give one, and keep each under 20 words. Bullets: Answered customer calls. Fixed billing problems. Trained new staff.
Why it works: It names the job, the style rules and a limit on inventing numbers, so the result stays truthful.
Code Explanation
Weak prompt: Explain this code.
Stronger prompt: Explain the following Python function to a beginner in plain language. First give a one-sentence summary, then walk through it line by line, then list two ways it could fail. Function: def total(prices): return sum(prices)
Why it works: It names the audience and breaks the answer into ordered steps.
Product Description
Weak prompt: Write a description for my candle.
Stronger prompt: Write a 60-word product description for a hand-poured soy candle called Cedar Morning. The audience is gift shoppers. Use a warm tone, mention a 40-hour burn time and end with a call to action. Give three versions so I can choose.
Why it works: It supplies the product facts, the audience, the tone, a length and asks for options.
Sorting Messages With Examples
Weak prompt: Sort these customer messages.
Stronger prompt: Sort each customer message into Billing, Shipping or Other. Example: My package has not arrived. Category: Shipping. Example: I was charged twice. Category: Billing. Example: Do you have a store in Austin? Category: Other. Now sort this message: Where is my refund?
Why it works: The examples show the exact pattern and format, which is how few-shot prompting works.
Free Prompt Builder
Step 1. What do you want the AI to do?
Step 2. Who will read the result?
Step 3. Tap the parts to include. Untick any you do not need.
Step 4. Click any text in the prompt and type to make it yours.
Replace the sample details with your own, then copy the prompt into your AI tool.
My Prompt
Copy this into your AI tool, then edit the details
- You are an experienced business writer.
- You are a careful research assistant.
- You are a patient tutor.
- Write a short email asking a colleague to move our meeting to Monday at 10 a.m.
- Summarize the article below in five bullet points.
- Create a four-week study plan for learning basic Spanish.
- The reader is a busy manager, so get to the point quickly.
- The reader is a beginner, so use plain words and explain any new terms.
- The reader is a customer, so keep a warm, helpful tone.
- Context: the meeting is with a client, and the reason for the change is a schedule conflict on Friday.
- Context: the article is about remote work trends, and I need the main points for a team meeting.
- Context: I can study 30 minutes a day, and I want to learn phrases for travel.
- Format: three short paragraphs with a clear subject line.
- Format: a bulleted list, with each bullet under 20 words.
- Format: a table with columns for week, daily focus and a small goal.
- Limits: keep it under 150 words, do not invent facts, and ask me a question if anything is unclear.
To save: press Ctrl+P (Windows) or Cmd+P (Mac) and choose Save as PDF. Only the prompt prints. To copy part of it, select the text and press Ctrl+C or Cmd+C.
The Parts of a Strong Prompt
| Part | What it does |
|---|---|
| Role | Sets the point of view |
| Task | States what to do |
| Context | Gives background |
| Audience | Sets tone and level |
| Format | Shapes the output |
| Limits | Sets length and rules |
You do not need every part every time. A short task may need only a task and a format. A bigger task benefits from a role, context and limits.
Techniques From the AI Makers’ Guides
- Be clear and direct. Anthropic suggests asking whether a colleague with minimal context would be confused by your prompt. Google’s Workspace tips say to be clear and concise and to use natural language.
- Give context and the reason. Anthropic says explaining why a behavior matters helps the model generalize. Google says context helps answers fit your situation.
- Show examples. Anthropic recommends 3 to 5 examples for best results. Google recommends examples but warns that too many can cause overfitting.
- Structure the prompt. Anthropic and OpenAI both suggest separating instructions, context and examples, for instance with tags or headers.
- Specify the format. Google advises naming the format, such as a table, a bulleted list or a paragraph. Anthropic suggests saying what to do instead of what not to do.
- Break big tasks into steps. Google recommends one prompt per instruction and chaining outputs from one step to the next.
- Iterate. Google suggests rephrasing, reordering and trying again when results miss the mark.
How to Practice
- Pick a real task. Use something from your work or studies, such as an email, a summary or a plan.
- Write a plain first prompt. Run it and read the result.
- Add one part at a time. Add a role, then context, then a format, and compare the answers.
- Keep a prompt log. Save the prompts that worked and what you changed.
- Test on different tools. The same prompt can give different results in different models.
- Check the output. Verify facts, numbers and quotes before you use the answer.
Limits and Safe Use
- AI can be wrong. A clear prompt helps, but it does not guarantee accuracy.
- Watch for bias. Read answers critically, especially for people, hiring or health topics.
- Protect private data. Do not paste confidential information into an AI tool unless your organization allows it.
- Settings vary. Google notes that some newer Gemini models ignore custom temperature values, so read the documentation for your tool.
- Tools change. Models and features are updated often, so retest your best prompts.
Mistakes to Avoid
- Being vague. Name the task, the audience and the format.
- Asking for too much at once. Split big requests into steps.
- Using one prompt style for everything. Change your wording to fit the task.
- Trusting the first answer. Review it, then refine the prompt.
- Typos in important text. Proofread twice, or use our free proofreading service.
Prompt Checklist
Tick each item as you finish. Your prompt is ready when the counter reaches 10.
Frequently Asked Questions
What is prompt engineering?
Prompt engineering is writing and refining the instructions you give a generative AI tool so the answer is more useful. It is mostly clear writing: say what you want, for whom, in what format and within what limits.
Do I need to know how to code?
No. For chatbots and writing tools you write in plain language. Developers who use an AI model through an API add settings, system messages and tests, but the basic skill is the same.
What makes a good prompt?
Google’s Workspace tips say to use natural language, be clear and concise, provide context, use specific keywords and break complex tasks into separate prompts. Claude’s guide suggests a test: if a colleague with minimal context would be confused by your prompt, the AI will be too.
How many examples should I include?
Anthropic recommends 3 to 5 examples for best results with Claude. Google recommends including examples but warns that too many can cause overfitting, and that all examples should use identical formatting. OpenAI suggests a handful of varied input and output pairs. Start with two or three.
Should I give the AI a role?
It often helps. Anthropic says even one sentence in the system prompt, such as telling the AI it is a helpful coding assistant specializing in Python, can focus its behavior and tone.
Why should I split a big task into steps?
Google’s Gemini guide recommends one prompt per instruction, and chaining steps so each output feeds the next prompt. Smaller steps are easier to check and fix.
Where should I put a long document in my prompt?
Anthropic recommends putting long documents near the top, above your question and instructions, and says that placing the query at the end improved response quality by up to 30% in its tests. Other tools may differ, so test with yours.
Does temperature matter?
Settings differ by tool and model. Google’s docs say some newer Gemini models ignore custom temperature values, so check the documentation for the tool you use before you rely on a setting.
Can AI still make mistakes with a good prompt?
Yes. A clear prompt improves your odds, but it does not guarantee accuracy. Check facts, names, numbers and quotes, watch for bias, and do not paste private or confidential information into a tool unless your organization allows it.
How do I get better at prompt engineering?
Practice on real tasks, change one thing at a time, and keep a log of prompts that worked. Read the official guides from the tool makers, and test again when a model is updated. Developers can also build evaluations, which OpenAI recommends running whenever you change or upgrade model versions.
Conclusion
Good prompts are clear, specific and tested. Start with a plain request, add the parts that matter, check the answer and save what works. Then use the builder above to write your next prompt.
Related Reading
- 10 best degrees for 2027
- 10 best degrees for the future 2030
- 6 cheapest online software engineering degrees
- Basic skills for resume
Sources: Anthropic, Prompting best practices; OpenAI, Prompt engineering; Google, Gemini API prompt design strategies; Google Workspace, Tips to write prompts for Gemini. Examples are fictional. AI tools and their guidance change often, so check the current documentation. This page does not guarantee any result from an AI tool.