Read this article on Linkedin: https://www.linkedin.com/pulse/engenharia-de-prompts-estruturando-intera%C3%A7%C3%B5es-com-modelos-sanchez-twxpf/?trackingId=l5E7gv5BIHprSQNS6R1mkA%3D%3D
The interaction between humans and artificial intelligence systems has evolved beyond traditional programming. Today, the real difference lies in how we formulate questions — an emerging field known as Prompt Engineering.
Prompt engineering is the point of convergence between human language and generative AI systems. It defines how instructions are translated into reasoning and structured responses, directly influencing the quality and accuracy of the output.
In practice, writing a good prompt is more like design an algorithm in natural language than just “asking a question.”
1. Clarity is more important than complexity
A common mistake is to assume that long instructions produce better results. The key is clarity of intention.
Bad example:
Write a text about solar energy, but in an interesting, educational, and perhaps a little technical way.
Better example:
Write a 300-word technical article explaining how photovoltaic solar panels work. Use engineering terms, include an explanation of efficiency, and conclude with a bullet-point summary.
Test on ChatGPT: Paste the two examples above and compare the difference in response structure and terminology.
2. Add context and role
Language models respond best when they know who are they representing and who are they talking to.
Technical example:
You're a senior data engineer. Explain to an intern how to build a secure and scalable ETL pipeline. Use simple analogies, highlight best practices, and address security risks.
This format — called role prompting — directs the model's thinking and sets the tone, level of depth, and technical vocabulary.
Test on ChatGPT: Try changing the role to “university professor” or “software architect” and see how the style changes.
3. Structure your reasoning into steps
Ask the model think step by step reduces errors and improves coherence, especially in analytical or decision-making tasks.
Example:
Explain step-by-step how to identify performance bottlenecks in a web application. List the most common tools for each step.
This technique is called chain-of-thought prompting (or chained reasoning).
Test on ChatGPT: Ask a complex task (e.g., optimizing SQL queries) and add the phrase "explain your reasoning step by step." Compare the clarity of the answer with and without this instruction.
4. Use examples (Few-Shot Prompting)
When the model needs to follow a specific format, offer input and output examples is the most effective way to guide the structure.
Example:
Example 1: Question: “How does supervised learning work?” Answer: “Supervised learning uses labeled data to train models that can predict outputs based on new inputs.”
Test on ChatGPT: Note how the model replicates the pattern of structure and technical language.
5. Control the output format
Define the desired response type—flowing text, table, JSON, bullet points, or code. This simplifies integration with automated workflows or post-processing tools.
Example:
Generate a table with three columns: Prompt Technique, Description, Practical Benefit. List at least five prompt engineering techniques.
Test on ChatGPT: Take a look at how the template organizes the content and try adding the "return in Markdown format" instruction to integrate with technical documentation editors.
6. Iterate as an engineering cycle
Prompt engineering follows a refinement cycle:
- Create the initial prompt
- Analyze the response
- Identify gaps
- Adjust instructions
- Rerun
Test on ChatGPT: Give a long assignment (e.g., “create a cybersecurity plan for small businesses”) and refine it with additional instructions until you reach the desired level of detail.
Conclusion
Prompt engineering is an emerging discipline that combines linguistic clarity, logical reasoning and interaction design. Mastering these techniques transforms the relationship with language models — from passive users to reasoning engineers.
In generative AI, the one who masters the prompt doesn't type—they design behavior.
CEO | Leading the AgenticAI Revolution for Enterprise
October 6, 2025
