AI advice
Writing prompts for beginners: How to get better answers
In short
A good prompt tells the AI what you want to achieve, who the result is for, what form you need it in and what it should avoid. The more context you give, the less the AI has to guess. Start with a clear task, read the answer critically and refine it in the same chat with targeted follow-up requests instead of starting over every time.
You ask an AI a question and get an answer that somehow fits but doesn't really help: too general, too long or in the wrong tone. That's rarely down to the tool and usually down to the instruction. A prompt is simply your brief to the AI, and just as with a new colleague, whoever says clearly what they need gets something better back. You don't have to become a pro for this. With six building blocks, one example and a few templates, you'll quickly get noticeably better results.
What goes into a good prompt?
A strong prompt answers the questions a person in your place would ask next. These six building blocks help you do that:
- Goal: What should come out at the end? “Write a friendly rejection” is clearer than “Help me with an email”.
- Context: What is it about, who will read the result, what do you already know? The AI doesn't know your situation.
- Role: From what perspective should the AI answer, for example “as an experienced tax adviser explaining something to laypeople”?
- Format: Length, structure and tone, for example “three bullet points”, “as a table” or “casual and informal”.
- Examples: A short sample of what the result should look like often works better than any description.
- Limits: What should the AI leave out or keep in mind? For example “no jargon”, “100 words maximum” or “don't make up any figures”.
Not every prompt needs all six building blocks. For a quick question, goal and context are enough. The more the result matters to you, the more building blocks are worth adding.
What does a better prompt look like in practice?
Before: “Write me a text for my website.” The AI knows neither what it's about nor who it's for. What you get is an interchangeable sales text full of empty phrases that you have to rewrite almost completely.
After: “I'm a self-employed photographer in Freiburg and mainly photograph weddings. Write a text for the homepage of my website. It's aimed at couples between 25 and 40 who prefer natural pictures to posed shots. Write like an experienced copywriter. Length: about 120 words, one headline and two short paragraphs, in a personal, informal tone. Take the style of this sentence as a guide: ‘I capture what you'll want to feel later.’ Avoid superlatives and don't make up any awards.”
The second prompt contains all six building blocks: goal, context, role, format, example and limits. It takes you one minute longer but saves several rounds of corrections.
Tip: Missing the context or not sure what the AI needs? Turn the tables and write: “First ask me five questions you need answered to write a good text.” That way the AI gathers the necessary information itself.
Follow up instead of starting over: How do I refine an answer?
The first answer is a draft, not a final result. Stay in the same chat, because there the AI knows your previous instructions. Then give concrete feedback instead of a vague “Make it better”:
- “Shorter, 80 words maximum.”
- “The second paragraph sounds too salesy, make it more factual.”
- “Give me three variations of the headline.”
- “What else do you need to make the result better?”
It's best to start a new chat when you change topic or the conversation has become very long. In long chats the AI can lose sight of early instructions, because its context window is limited. In that case, first ask it to summarise all your instructions and the result so far in one paragraph, then copy that into the new chat.
Which prompt templates help with typical tasks?
You can copy these templates straight away. Just replace the details in square brackets with your own.
- Email: “Write a short, friendly email to [recipient] in which I [request]. Tone: [casual or formal]. 120 words maximum.”
- Summarise: “Summarise the following text in five bullet points for [target audience]. List any open questions at the end. Text: [paste]”
- Explain: “Explain [topic] to me as if I had [prior knowledge]. Use an everyday example and no more than 150 words.”
- Brainstorm: “Give me ten ideas for [goal]. Sort them by effort, from easy to hard.”
- Improve a text: “Revise this text: easier to understand, shorter sentences, same content. Then list what you changed. Text: [paste]”
- Decide: “I'm torn between [A] and [B]. [Criterion] matters to me. List the pros and cons and give a reasoned recommendation.”
How several prompts work together across a whole project is shown in our guides, for example the guide for a blog article. Before you paste in texts containing customer or health data, take a quick look at what you should and shouldn't enter.
What mistakes do beginners make most often when prompting?
- Too short, like a Google search: Three keywords are enough for a search engine, but an AI fills the gaps with generic statements.
- Everything at once: It's better to work step by step, first the outline, then the text, then the shortening.
- No format specified: Then the AI decides on length and structure, and the result is often too long.
- Only saying what's wrong: “Not so formal” helps less than “Write the way you'd talk to a colleague”.
- Accepting answers without checking: AIs phrase mistakes convincingly too, so-called hallucinations. That's why you always check figures, quotes and sources yourself.
Tip: Our recommendation: Save prompts that worked well in a simple note. That way you gradually build up your own collection of templates and don't have to start from scratch every time. How to cross-check answers properly is shown in the guide Checking AI answers.
When is a different AI or tool the better choice?
Sometimes it's not the prompt but the tool. All-round assistants like ChatGPT, Claude or Google Gemini are strong generalists, but not the best choice for every task.
- Current facts with evidence: A research tool like Perplexity shows you where the information comes from. More on this in the tools for research and knowledge.
- Translations: Specialists like DeepL (Free) or Google Translate are designed for exactly this, see translation tools.
- Images: For these you need an image generator. Prompts there mainly describe subject, style, lighting and perspective, more in the category Image generation.
- Your own documents: Gemini Notebook answers based on the sources you upload instead of from general knowledge.
If an assistant gets stuck despite a good prompt, try the same prompt in a second one. The models have different strengths, and the comparison only costs you a minute. Which assistant suits you in general is covered in the guide ChatGPT, Claude, Gemini or Mistral compared.
Frequently asked questions
How long should a good prompt be?
There's no fixed length. For a simple question one sentence is enough; for a text that will be published, more like a short paragraph with goal, context and format. What matters is not the word count but whether the AI knows everything it needs for a good answer. If something is missing, you'll notice it in general or evasive answers.
Do I have to be polite to the AI to get a better answer?
Politeness doesn't hurt and makes writing more pleasant. Clear details on goal, context and format make the bigger difference, though. A friendly “please” doesn't replace missing information. So feel free to write the way you'd talk to a person, and focus above all on saying clearly what you need.
Do the same prompts work in ChatGPT, Claude and Gemini?
Yes, the basic rules apply to all the major assistants. Goal, context, format and limits improve the answers everywhere. The results still differ, because each model has its own strengths and its own writing style. If you don't like an answer, it's worth simply testing the same prompt in another assistant.
What's the difference between a prompt and prompt engineering?
A prompt is the single instruction you give the AI. Prompt engineering refers to systematically developing and testing such instructions, often for recurring tasks or programs. As a beginner you don't need any expert tricks: if you use the six building blocks from this guide and ask targeted follow-up questions, you'll already get very good results in everyday use.