Key principles for creating an effective prompt

When you ask an AI tool – whether it is a conversational system or an image-generating tool – a query (called a prompt), it does not interpret it as a human would. These systems do not possess consciousness, thought, or understanding in the human sense. Instead, they use advanced mathematical models and statistical patterns to identify what you are seeking and generate the response that best matches your query.

Here is how it works:

• Converting to numbers

  • The first step is to convert your query – whether text, image, or sound – into a series of numbers. This is how a computer “translates” information into its own language.
  • If the input is text, it is broken down into smaller units (words or parts of words), each assigned a numerical value that reflects its meaning and its relationship to other words.
  • If it is an image, the data is converted into numbers that describe the characteristics of its pixels.

• Finding patterns and connections

         Once all information is expressed as numbers, the AI tool begins its analysis. It uses the extensive “knowledge” acquired during training on billions of examples.

• • Recognising relationships: The model identifies how different parts of your query relate to each other. For example, in the phrase “a dog that flies,” it understands that the verb “flies” describes the noun “dog.”

  • Focusing on the essential: Like a reader searching for the main idea of a text, the system determines which words or parts of the query are most important and focuses on them.

  • Comparison with learned patterns: Your query is compared to what the model has “learned” during training on large data sets. The model does not remember individual examples but generalises patterns, language structure, facts, and ways to connect information.

• Generating answers

         Based on this analysis, the AI tool predicts the most likely or appropriate answer to your query.

• • For text queries: The model predicts the next most likely word, then the next, gradually building the entire answer.

  • For visual queries: It gradually “draws” an image from numerical data, pixel by pixel, creating a representation that matches the description in the query.

Artificial intelligence tools do not think or understand the world like humans. Their strength lies in recognising complex patterns and making statistical predictions. This is why they can create extremely convincing and useful answers – thanks to sophisticated data processing and the ability to recognise connections that humans have enabled during the learning process.

What is prompt engineering?

Prompt engineering is the skill and process of designing effective queries for artificial intelligence tools.

It is a new form of digital literacy that involves understanding how artificial intelligence systems “think” and respond to different formulations of requests.

The goal of prompt engineering is to maximise the capabilities of artificial intelligence through clearly defined, contextually precise, and purposefully structured queries. In practice, this means the user carefully chooses:

• Role – from which perspective the tool should respond

• Task – what exactly needs to be done

• Format – in which format the result should be displayed

• Restrictions – rules, tone, or response limits

Example of a prompt:

“You are a math teacher. Prepare a short lesson plan on linear equations for third-year high school students. Display the plan in a table and use simple, clear language.”

This approach allows users to precisely direct the work of artificial intelligence, increase the relevance of responses, and develop awareness of how digital tools process information.

In education, prompt engineering is becoming an important skill for teachers and students, as it enables the conscious and responsible use of artificial intelligence in learning, research, and content creation.

Creating different types of content (e.g., text, visual, or audio) often requires different artificial intelligence tools and different query design strategies. In other words, how we formulate a prompt depends on what we want to produce: text, image, sound, or multimedia.

Specific rules and approaches apply to each type of content.

Here is what to pay attention to for certain types of prompts:

Textual content – requires a clearly defined tone, goal, and structure. It is important to determine the intended audience, the style (e.g., formal, popular, educational), and the specific content required (e.g., summary, explanation, comparison).

Visual content – requires precise descriptions of image elements, style, composition, and mood. The general rule is: the more detailed the description, the more precise the result.

Audio and audiovisual content – includes additional dimensions such as tone of voice, rhythm, emotion, or atmosphere, making queries more complex.

Multimodal content – combines text, image, and sound, and requires carefully coordinated instructions for the tool to properly integrate all components.

Understanding these differences helps users choose the appropriate AI tools and methods of interaction, thereby increasing the accuracy, creativity, and efficiency of the results.

Additionally, for those seeking a deeper understanding of prompt engineering strategies and techniques when working with large language models, the Prompt Engineering Guide developed by OpenAI is recommended.
The guide provides examples, practical advice, and advanced approaches to query design to improve the quality and consistency of generative model responses.

Note: The presented query design model (Role–Task–Format–Restrictions) is based on pedagogical guidelines for developing digital and AI literacy used by various university guides, such as those from the Australian Catholic University (ACU) (2024). Although not formalised as a theoretical model in the scientific literature, this approach serves as a practical framework for training users in effective communication with AI tools.

Image source: Shutterstock