If you've used ChatGPT, Gemini, or another AI before, you've probably experienced this: you ask a question and the answer is… strange. Neither entirely bad, nor truly useful. And that's where the doubt arises: Is AI useless, or am I asking the wrong question?
Spoiler alert: it's almost always the latter. This is where the key concept of this article comes into play: the prompt.
What is a prompt in AI?
A prompt is the instruction you give to artificial intelligence to generate a response. It can be a short sentence, a long paragraph, a direct question, or a combination of context and task.
The difference between a regular one and a powerful one lies in how do you phrase it?.
Why most prompts fail
Here's an important truth: AI doesn't "make mistakes", it interprets what you give it.
- If the prompt is ambiguous, the response will be ambiguous.
- If you don't define the objective, the AI improvises.
- If you ask for "a little bit of everything", you'll get... a little bit of nothing.
That's why two people can use the same tool and get completely different results. The key isn't in the AI, but in... the quality of the prompt.
Structure for a good prompt
Without getting too technical, a good prompt usually has four basic ingredients:
- RoleFrom what perspective should it respond?
- ContextWhy do you need the information?
- InstructionWhat exactly should you do?
- FormatHow do you want the result?
You don't always need all of them, but the more complex the task, the more important this structure becomes.
Think of the prompt as an experiment: if the design is well thought out, the results will be better.
Prompt templates you can use today (by profile)
Here are practical examples adapted to scientific and professional profiles.
1. Research – Literature Review
I work as a researcher in [discipline]. I need a synthetic review on [specific topic], focusing on studies published in the last 5 years. Summarize the main approaches, key results, and methodological limitations. Include suggested references to verify the information and clarify when something is an estimate.
2. Teaching – Structured lesson plan
Act as a university lecturer for [subject]. Design a 90-minute lesson plan on [topic] for [undergraduate/master's] level students. Include learning objectives, a timeline, a practical activity, and assessment criteria. Present it in table format.
Here you are already guiding the format, level and actual use.
3. Analysis – Create operational checklist
Analyze the following procedure and create an operational checklist to detect common errors. Organize your response in a table with the following columns: Step, Risk, Failure Indicator, Corrective Action. If any data is missing, indicate this explicitly.
This turns AI into a methodological assistant, not a generator of pretty text.
4. Scientific writing – Critical summary of paper
Summarize the following scientific article in 250 words. Include: objective, methodology, main results, and limitations. Point out any possible biases or implicit assumptions. Do not invent data that does not appear in the text.
That last line is a key guardrail.
5. Improvement of technical text
Rewrite this text to make it clearer and more precise, while maintaining scientific rigor. Reduce ambiguities and avoid absolute statements without evidence.
Ideal for articles or reports.
6. Hypothesis generation
Based on these preliminary results, generate three plausible alternative hypotheses. Explain the assumptions of each and what experiment would validate it.
This is already assisted structured thinking.
Guardrails: how to avoid unreliable answers
Very important: AI can generate incorrect information with a high degree of certainty.
Always add instructions such as:
- “Include sources or indicate if you are unsure.”
- “Separate facts from hypotheses.”
- “List implicit assumptions.”
“"It indicates limitations of the analysis."” - “Don’t make up references.”
And then, verify. AI helps you think faster, but the final responsibility remains yours.
In science and professional work, this is non-negotiable.
Common mistakes when writing prompts
Some very common mistakes (and totally normal when starting out):
- Asking vague questions: “Tell me about…”
- Do not indicate what the answer will be used for
- Expecting AI to "guess" the technical level
- Asking for too many things in one sentence
- Do not review or improve the prompt after the first response
The good news is that All these errors are easily corrected with a little practice.
How to improve your prompts step by step
The first prompt is almost never the final one.
Practical process:
- Write a clear first draft.
- Evaluate the response.
- Adjust: add restrictions, change format, limit scope.
- Repeat.
The improvement is incremental.
A good prompt is rarely born perfect. It is perfected.
Checklist to evaluate your prompt
Before sending it, ask yourself:
- Have I defined the objective?
- Is the technical level clear?
- Have I limited the scope?
- Have I requested a specific format?
- Have I included verification criteria?
If you answer “yes” to most of them, you’re on the right track.
From generic AI to mastering prompting
Mastering prompting is a game-changer.. It's not just about using AI and that's it; almost everyone is already doing that. It's about managing and controlling how you talk to your AI so that it gives you what you want.
It's not about AI thinking for you, but about Do better what you already know how to do.Check out our course and learn how to create prompts Perfect. If AI is your ally, you'll appreciate this training.
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.
