Written by Alberto Fernández
Prompt engineering is the art of communicating effectively with AI models to obtain precise and useful responses. In essence, it is about choosing the right words, context, and format to guide the model’s behavior toward the result you need.
The way you phrase a question greatly influences the quality of the response. The same query posed in different ways can generate completely different results, demonstrating why it is so important to fine-tune your instructions.
With the rise of language models, a new specialty has emerged: the prompt engineer. These professionals focus on finding the best ways to interact with AI—not by programming it directly, but by teaching it to better understand what we want from it.
In many projects, the challenge isn’t whether the AI can generate text, but getting it to generate the right text in the right way. A model may have extensive knowledge, but a vague or poorly focused query will produce generic, incomplete, or incorrect answers.
This is where prompt engineering comes into play. With well-designed instructions, you provide the model with the necessary context and guidelines so that its response is useful and aligned with your needs.
For example, asking “Explain the importance of Paris” is not the same as specifying: “Briefly explain the historical importance of Paris, using a formal tone and mentioning its cultural aspects.” This second version guides the model on what to answer and how to do it, significantly increasing the quality of the result.
As you can see, any change in phrasing can have a major impact on the final outcome.
How to write effective prompts
Although prompt engineering is a complex discipline, there are fundamental practices you can apply immediately to improve your results with AI.
Clarity and context: the foundation of everything
An effective prompt minimizes ambiguity. Include relevant details: what information you need, what the focus should be, and who the audience is. The more descriptive your instruction, the better the AI will understand your intent.
For example, instead of “Summarize the document,” specify: “Summarize the conclusions of the following technical document on solar energy in one paragraph.” If a term can be interpreted in multiple ways, clarify it to avoid misunderstandings.
Define the response format
Explicitly indicate how you want to receive the information. By default, AI responds in plain text, but if you need a list, a table, code, or JSON, specify it clearly. For example: “Provide the response as a bulleted list.”
Defining the format saves time and gives the model a clear presentation guide.
Break down complex tasks
If your request is very broad, break it down into manageable subtasks. Asking for too much at once often leads to superficial responses.
It is much more effective to maintain a multi-turn dialogue, where each prompt addresses a specific aspect. For example, first ask “Generate a list of key features for X,” then “Using those features, write an introduction,” and so on.
Assign a role to the model
Telling the AI what role it should assume or what tone to use significantly improves the quality of the result. For example: “Act as an expert SEO consultant and explain…” or “Respond in a professional yet accessible tone…”.
Focus on what you DO want
Tell the model what to do instead of just listing prohibitions. Instructions phrased only in the negative can confuse the model.
Instead of “Don’t give a short answer,” indicate “Provide a detailed response of at least 4 paragraphs.” The results will be much better.
Iterate and experiment
The first response is rarely perfect, but that’s normal. Prompt engineering requires experimentation. If you’re not satisfied with the result, adjust the question, add more context, and try again.
Continuous iteration is key to refining your prompts and achieving optimal results.
Use examples (few-shot prompting)
Providing the model with concrete examples of what you expect is extraordinarily effective. AI responds better when it has a model to follow.
If you want a specific headline, include a well-written example before your request. Provide one or two high-quality, relevant examples, but don’t overdo it—too many examples can confuse the model.
Request step-by-step reasoning
For complex tasks that require logic, ask the model to reason step-by-step. This technique (chain-of-thought) involves instructing it to show its logical process before reaching the final conclusion.
For example: “Think through each step before responding” or “Explain your reasoning.” This makes the model pay more attention to breaking down the problem, reducing errors.
Keep in mind that this instruction increases tokens and cost, so use it only when accuracy is critical.
Prevent hallucinations
Even with good prompts, there is a risk that the AI will invent information. To mitigate this, include instructions on how to handle uncertainty.
For example: “If any information is unavailable, state that you do not know it instead of making it up.” This sets boundaries on the model’s creativity where you need reliability.
For topics requiring up-to-date data, provide the recent context yourself and require the model to limit itself to that evidence.
Advanced formatting techniques
A well-structured prompt gives the AI clear clues on how to respond. Let’s look at some professional techniques:
Delimiters and clear structure
Use symbols or tags to delimit different parts of the prompt and avoid confusion. These can be quotes, brackets, or symbols that clearly indicate where each piece of information begins and ends.
For example, if you want to translate a text, isolate it with quotes. You can also divide complex prompts with subheadings like ### Instruction ###, ### Context ###, and ### Question ### to differentiate each section.
Structured prompts (JSON/XML)
An advanced technique involves providing the request in JSON or XML format instead of natural language. Models have seen a lot of structured code during their training and tend to respect it.
For example, instead of “Summarize the customer’s opinion on the shipping,” you could use:
{
“task”: “summarize”,
“topic”: “customer_opinion”,
“focus”: “shipping”
}
The main advantage is that it reduces ambiguity and can force responses into a predictable format, which is especially useful when you need to process the output automatically.
However, writing the prompt in JSON does not guarantee better responses on its own. It primarily adds value when you need structured outputs or format validation.
Emphasis with capitalization
Highlighting important instructions with capitalization can help them stand out. For example: “You MUST respond only with verified information.”
Using all caps is useful for separating key elements, but remember that:
- It does not guarantee absolute obedience
- It does not automatically prioritize that instruction over others
- On its own, it is not enough
Conclusion
Prompt engineering has become an essential skill for making the most of the potential of AI models. Mastering this discipline does not require programming knowledge, but rather developing a strategic mindset on how to communicate effectively with these systems.
The techniques we have explored—from clarity and context to the use of examples and advanced structuring—are tools that anyone can apply immediately. The key lies in experimenting, iterating, and learning from each interaction.
As AI models continue to evolve, the ability to formulate effective prompts becomes increasingly valuable. It is not just about getting answers, but about getting the right answers, tailored to your specific needs and presented in the most useful way possible.
Remember that prompt engineering is both a science and an art. Apply these fundamental practices, but don’t be afraid to experiment and develop your own style. With time and practice, you will achieve increasingly precise and valuable results in your interactions with AI.
