4 minutes to read

Prompt Engineering, from art to science?.

Hylenne González

The world of Artificial Intelligence has undergone an unprecedented boom in the last year and prompt engineering has become one of its most intriguing areas. But what is it exactly? And why should anyone in tech or analytics care? This article will help you figure it out:

The concept

Prompt Engineering refers to the practice of designing and fine-tuning the prompts (or instructions/requests in Spanish) given to the Large Language Model (LLM), such as Open AI‘s ChatGPT, with the goal of obtaining accurate, relevant or useful responses. of these systems. Given the generalist nature of many advanced language models, the way a question or statement is phrased can have a significant impact on the quality and usefulness of the response generated.

It is both an art and a science, as it requires a combination of technical understanding of the LLM and creativity to design prompts that produce the desired results.

The origin

From the first chatbots until now, we have witnessed a notable evolution, but with the arrival of mass-use models like ChatGPT, it has become evident that the way we “request” information from these models—or how we formulate our “prompts”— it’s crucial.

By the way, do you know that there are other LLMs besides ChatGPT? In our post Guide to understanding and selecting Generative Artificial Intelligence tools we tell you about four other options that are within our reach.

Let’s go back to Prompt Engineering…

You may be wondering, what do I get by applying Prompt Engineering? And the truth is that from the automatic processing of customer feedback to the generation of executive summaries, the possibilities are vast, which is why we present some of them below:

  • Improved responses: Through careful design of prompts, more accurate and relevant responses can be obtained.
  • Operational efficiency: Reduces the time and effort required to extract information or perform specific tasks with language models.
  • Reduced costs: Minimizes the need for repeated manual interventions by getting the right answers right from the start.
  • Flexibility: allows general language models to be adapted to specific applications or market niches without having to train models from scratch.
  • Insight discovery: With well-designed prompts, you can unravel valuable information from large data sets or texts.
  • Continuous improvement: As prompts are refined, based on feedback and results, the quality and relevance of responses tend to improve.
  • Advanced Automation: Makes it easier to create smarter, more contextual automated systems for tasks like customer support and data analysis, for example.
  • Resource optimization: By maximizing the utility of powerful language models, you get the most out of your investment in AI technologies.
  • Adaptability: allows companies and developers to respond quickly to new needs or changes by reconfiguring prompts, without needing to alter underlying models.
  • Innovation: Exploring different ways of interacting with language models opens doors to new approaches and solutions to existing or emerging problems.

Best practices

Below are ten recommendations to implement effective Prompt Engineering in your day-to-day use of Generative Artificial Intelligence:

  1. Understand the Model: Familiarize yourself with the capabilities and limitations of the language model you are using.
  2. Clarity, specificity and context: formulate clear, specific and contextualized prompts to reduce ambiguities and obtain more precise responses.
  3. Iteration and testing: design several prompts for the same objective and test which one produces the most appropriate results.
  4. Use of examples: provide concrete examples within the prompt to guide the model towards the desired answer (step by step, enumeration, list, table, short sentences, descriptive format, imaginative format, etc.)
  5. Avoid bias: Be aware of any bias in your prompt and avoid words or phrases that could bias the model’s response, for example: qualifying adjectives.
  6. Length control: experiment with the length of the prompt. Sometimes more detailed prompts are useful; other times, brevity is more effective.
  7. Using open and closed questions: Depending on what you’re looking for, decide whether it’s best to ask an open question (for descriptive answers) or a closed question (for concise answers).
  8. Continuous Feedback: Use feedback from users and other systems to refine and improve your prompts over time and use.
  9. Documentation and logging: Keep track of which prompts work well and which ones don’t, so you can learn, adapt, and save time in the future.
  10. Integration with Other tools: Consider how prompts and responses integrate with other tools or systems in your workflow. Make sure the answers are compatible and useful for the intended applications.

Of course, in all recommended practices the sine qua non condition of being critical of the responses that the model used generates applies.

From the theory to the practice

These examples demonstrate how an optimized prompt can direct the language model to provide more structured, specific, and useful responses:


Prompt Engineering emerges as an essential skill in the universe of Generative Artificial Intelligence. Mastering it can transform the way we interact with language models, allowing us to unleash their maximum potential and, consequently, provide value in decision making and data analysis. The next time you find yourself in front of ChatGPT or any other LLM, remember that how you “ask” is as essential as what you are looking for.

At Mottum, we understand the importance of implementing effective AI strategies in your projects. That’s why we’re offering you a free strategy session where we’ll explore how Prompt Engineering can adapt and enhance your specific initiatives. Book your free session now!