AI chat assistants can help with many tasks, from writing and summarising to planning and learning a new subject. But the quality of the answer depends heavily on how the question is asked. These questions are called "prompts". Here are seven rules for better results.
1. Give context
Instead of "Write an email", saying "Write a short, professional email to a customer telling them their order will be two days late, apologising without overdoing it" gives a far more useful result. Stating who it is for, why and in what situation reduces the assistant's need to guess.
2. Define the format
Do you want a bulleted list, a table or a three-paragraph summary? Clearly stating length and format reduces the need for edits later.
3. Show an example
If you want a particular style, sharing an example you like is the most effective approach. Rather than saying "write in this tone", adding a short text that shows the tone gives much clearer direction.
4. Break big tasks into parts
For a long report or complex project, ask for an outline first, then develop the sections one by one. Moving step by step improves quality and helps you spot mistakes early.
5. Have a conversation
The first answer doesn't have to be perfect. You can improve the result step by step with feedback such as "make it shorter", "explain the second point with an example" or "use a more formal tone".
6. Don't forget to verify
AI assistants can give fluent, confident answers, but they sometimes produce incorrect or invented information. Check important details against reliable sources, especially dates, figures, quotations and topics such as health, law and finance.
7. Don't share sensitive information
Think twice before pasting passwords, ID numbers, bank details or confidential company documents into chat tools. Reviewing the privacy settings and data policy of the service you use is a good habit.
A short template
The skeleton of a good prompt can be summed up as: role (for example "as an experienced editor"), task (what to do), context (for whom and why), format (length, structure) and constraints (what to avoid). A prompt with these five elements usually gives satisfying results even on the first try.
AI is not a tool that thinks for us; it is an assistant that speeds up and broadens our thinking. The final decision and responsibility always rest with the user.
Common mistakes and simple fixes
- One-word requests: A request such as “write a CV” brings a generic, template-like result. Fix: describe your experience, the job you are applying for and the length you want.
- Copying the answer as it is: The model can produce fluent but wrong sentences. Fix: reread the text in your own voice and edit it.
- Not questioning figures and sources: Models can produce sources that do not exist or wrong numbers. Fix: verify every important piece of information against an official source.
- Losing track in a long chat: In a conversation that runs very long, earlier instructions can be missed. Fix: start a new task in a new chat and briefly restate the key information.
What these mistakes have in common is treating AI like a search engine or a source of definitive facts; it is more accurate to think of it as an assistant that works fast but needs checking.
Responsible use at school and at work
UNESCO's guidance on generative AI in education, published in 2023, stresses that these tools should support learning but not replace human thinking. The US National Institute of Standards and Technology (NIST) also published a risk management profile for generative AI in 2024, which addresses risks such as fabricated information, privacy and copyright one by one. In practice this means: learn your school's or workplace's AI rules, say that you used a tool where this is required, and do not upload documents containing customer or personal data to tools that have not been approved.
Sources and further reading
- UNESCO (2023) – Guidance for generative AI in education and research
- NIST AI 600-1 (2024) – Generative Artificial Intelligence Profile
Leave a comment
Your email address will not be published. Required fields are marked *