If you've already started working with AI, you've probably experienced this: you ask it a question, the answer not bad, but It's not what you needed either.. And you think: “"Okay, I'll rephrase it."”. And then again. And again.
The good news is that It's not a matter of luck, but of structure. A well-designed prompt functions like a good experimental protocol: orderly, clear, and with a defined objective.
Let's see how to structure a prompt so that the AI understands exactly what you expect from it.
Key elements of a well-designed prompt
A good prompt doesn't have to be long, but it does well organized. These are the basic building blocks that work very well.
Role
Defining a role helps a lot. It's not the same to ask for a response as a science communicator, as a researcher, or as a teacher.
Examples:
- “Act like a biomedical researcher…”
- “Answer like a reviewer of scientific articles…”
- “Explain it as if you were a university professor…”
This It sets the tone, the technical level, and the approach from the beginning.
Context
AI doesn't know what you're doing or why you need it... unless you tell it.
Here's some key information:
- Where are you at in the project?
- What are you going to use the answer for?
- Who is it aimed at?
Example:
“I am preparing a literature review on machine learning applied to health…”
With just one sentence, the quality of the response usually improves significantly.
Instruction
This is the heart of the prompt.
It's not appropriate to be poetic here: Better clear than pretty.
Some useful verbs:
- Summary
- Analyze
- Compare
- Explain
- Extract
- Rewrite
- Propose
Example:
“Summarize the main findings and limit the explanation to the key points.”
The more specific the instruction, the less room there will be for answers that are not useful to you.
Output format
This point is often forgotten… and it makes all the difference.
Do you want it in long text?
In comic strips?
On a table?
In numbered steps?
Example:
“"Return the answer in a list of 5 points, with short phrases."”
This saves time and avoids having to ask for corrections later.
Complete example of a well-structured prompt
Let's put everything together in a single prompt:
“Act like an expert data science researcher.
I am preparing a literature review on the use of AI in medical diagnosis.
Summarize this article, highlighting its objective, methodology, and main results.
Present the information in a clear and concise list, using academic language.”
Do you see the difference?
It's no longer than other prompts… but it's much better organized.
One last important point
Mastering the structure of prompts isn't about controlling AI, but about communicate better with her. The clearer you are, the better the results you get. If you want to delve deeper and learn how to formulate prompts strategically (especially for research, scientific writing, or advanced analysis), there are specialized resources that teach you how. the art of asking the right questions, something that makes a real difference in everyday life.
Biotechnologist and biostatistician specializing in data analysis and artificial intelligence applied to research. Founder of Maxymia, combines science and technology to develop solutions and training in automation for the scientific field.
