Mastering ChatGPT Prompt Engineering: A Comprehensive Guide to Strategies and Techniques

In the fascinating world of artificial intelligence (AI), effective communication between humans and machines is key to unlocking the full potential of emerging technologies. One of the most powerful tools at our disposal is ChatGPT, a language model developed by OpenAI, which has revolutionized how we interact with AI systems. However, to maximize the effectiveness of these interactions, a critical skill is required: prompt engineering.

Prompt engineering is more than just asking questions; it’s an art and science that involves the careful formulation of instructions or queries directed at language models like ChatGPT, Microsoft Copilot, and Google Gemini, for instance. The goal is to optimize the quality, accuracy, and relevance of the responses. With the right approach, it’s possible to extract detailed information, generate creative content, or even guide the model to perform complex tasks with astonishing efficiency.

This article is a comprehensive guide that unveils the universe of prompt engineering, from basic principles to advanced techniques. Through practical examples and detailed strategies, you’ll learn to communicate more effectively with ChatGPT, turning simple interactions into rich and productive dialogues. Whether you’re looking to optimize your interactions with AI, explore new avenues of creativity, or simply gain a better understanding of how to talk to machines, you’re in the right place.

What is Prompt Engineering?

Prompt engineering is at the heart of effective interaction between humans and advanced language models like ChatGPT. It involves the careful creation of instructions or questions for AI models, aiming to optimize the quality and relevance of the responses. It’s an essential component for extracting the full potential of linguistic models.

Definition and Importance

Prompt engineering refers to the meticulous process of crafting instructions or questions directed at a language model to induce specific, informative, and relevant responses. This process goes beyond merely asking questions; it’s a strategic approach that seeks to fully exploit the model’s capabilities, ensuring that the interaction is as fruitful as possible.

The importance of prompt engineering lies in its ability to transform generic interactions into meaningful and goal-oriented conversations. In a world where AI is becoming increasingly integrated into our daily lives, being able to effectively communicate with these systems is not just an advantage; it’s a necessity.

A prompt is essentially a question or command you give to ChatGPT, expecting it to respond accordingly. For example, if you want ChatGPT to generate a poem about love, you might use a prompt like this:

Write a poem about love with four verses and rhymes.

ChatGPT might respond with something like:

Through lifetimes and stardust, love transcends,
A symphony written in heartbeats' blends.
In the garden of hearts, eternity weaves,
Love's fragrant petals, whispering leaves.

However, ChatGPT might not always respond exactly as you wish. Sometimes, it can generate irrelevant, incoherent, contradictory, offensive, or even dangerous responses. Therefore, it’s crucial to use prompt engineering to create queries that are clear, specific, contextual, creative, and experimental. And it’s equally important to review the responses.

Basic Principles of Prompt Engineering

Prompt engineering is a skill that requires practice and experimentation. There is no magic formula that guarantees ChatGPT will generate the perfect output for any prompt. However, there are some basic principles that can help improve the quality and relevance of ChatGPT’s responses. They are:

Clarity and Conciseness

Simplicity leads to more direct and useful responses. One of the fundamental principles of prompt engineering is clarity. Prompts should be clear and straightforward, avoiding ambiguities that could lead to inaccurate or out-of-context responses. Conciseness, in turn, helps keep the focus of the response, avoiding unnecessary digressions.

For example, if you want ChatGPT to generate Python code that calculates the average of a list of numbers, you might use a prompt like this:

Write Python code that takes a list of numbers as input and returns the arithmetic mean of the numbers as output.

ChatGPT might respond with something like:

def average(list):
  sum = 0
  for number in list:
    sum = sum + number
  mean = sum / len(list)
  return mean

Specific Intent

Clearly defining the prompt’s objective helps guide the responses. The prompt should be specific and detailed, leaving no room for varied interpretations. ChatGPT can stray off topic if the prompt is too broad or superficial.

For example, if you want ChatGPT to generate a summary of a book, you might use a prompt like this:

Write a summary of up to 200 words for the book "The Little Prince" by Antoine de Saint-Exupéry, highlighting the main characters, settings, and themes.

ChatGPT might respond with something like:

"The Little Prince" is a book by Antoine de Saint-Exupéry, published in 1943, that tells the story of a pilot who crashes in the Sahara desert and meets a boy from another planet, the little prince. The book is a masterpiece of children's literature, but it also delights readers of all ages with its messages of love, friendship, responsibility, and the meaning of life.

The little prince lives on a small planet, asteroid B-612, where he cares for a unique and special rose. He decides to travel through the universe, visiting other planets, where he meets strange and lonely characters, such as a king, a vain man, a drunkard, a lamplighter, a geographer, and a businessman. On each planet, he learns a lesson about human nature and his own.

On Earth, he meets a fox, who teaches him the meaning of taming and saying goodbye. He also meets the pilot, who becomes his friend and confidant. Together, they experience adventures and reflections, until the little prince decides to return to his planet, leaving an indelible mark on the pilot's heart.

Contextualization

Providing appropriate context improves the accuracy of responses. The prompt should offer the necessary context for ChatGPT to understand the purpose and scope of the query. ChatGPT can rely on the context to generate more appropriate and consistent responses.

For example, if you want ChatGPT to provide guidance and tips on how to specialize in order to work in the field of machine learning, you could use a prompt like this:

Considering an IT professional with five years of experience in software development and a growing interest in artificial intelligence, which career path would you recommend for specializing in machine learning?

ChatGPT might respond with something like:

For an IT professional with five years of experience in software development and growing interest in artificial intelligence, specializing in machine learning can open up many exciting opportunities. Here are some steps you can take to specialize in this area:

1. Learn the Fundamentals: Before diving deep into machine learning, it's important to have a solid understanding of fundamental concepts in mathematics, statistics, and programming. This includes linear algebra, calculus, probability and statistics, as well as programming skills in languages like Python and popular libraries like NumPy, Pandas, and Matplotlib.

2. Online Course or Certification: Consider taking online courses or obtaining certifications in machine learning. Platforms like Coursera, Udacity, edX, and Khan Academy offer a variety of courses in machine learning taught by industry and academic experts. Recognized certifications can help establish your credibility in the field.

3. Study Machine Learning: Dedicate time to studying the principles and algorithms of machine learning. Start with basic supervised and unsupervised learning algorithms, such as linear regression, classification, decision trees, k-means, etc. Then progress to more advanced techniques like neural networks, deep learning, natural language processing (NLP), and others.

4. Practical Projects: Apply your knowledge to practical projects. Start with simple projects and move on to more complex challenges as you gain experience. This can include data classification problems, time series forecasting, pattern recognition, and more. Building a machine learning project portfolio will demonstrate your skills to potential employers.

5. Join Communities: Join online machine learning communities like Kaggle, where you can participate in competitions, collaborate on projects, and learn from other professionals in the field. Also, engage in discussion groups, forums, and conferences to stay up-to-date with the latest trends and developments in machine learning.

6. Postgraduate or Specialization: If you are interested in further academic study, consider pursuing a postgraduate degree in artificial intelligence, machine learning, or related fields. A master's or doctoral degree in machine learning can provide a solid theoretical foundation and research opportunities.

7. Seek Professional Opportunities: Look for job opportunities that allow you to apply your machine learning knowledge. This might include positions such as data scientist, machine learning engineer, machine learning developer, and others. Consider also companies that are investing in AI and machine learning initiatives.

8. Stay Updated: The field of machine learning is constantly evolving, so it's important to stay updated with the latest research, technologies, and best practices. Read academic papers, expert blogs, books, and take refresher courses to ensure you're always at the forefront of the field.

By following these steps, you will be on the right track to specialize in machine learning and build an exciting career in this rapidly growing field.

The basic principles we’ve just covered are the main ones: clarity and conciseness, specific intent, and contextualization. They are the most cited, studied, and applied. But we can go a bit further. That’s why we present two more: creativity and experimentation.

Creativity

The prompt can be creative and original, without copying or imitating other prompts. ChatGPT can be inspired by the prompt to generate more interesting and fun responses.

For example, if you want ChatGPT to generate a funny tweet about the weather, you might use a prompt like this:

Write a funny tweet about the weather, using a pun or a joke.

ChatGPT might respond with something like:

It's so cold today that even my coffee froze. Now I have to drink iced coffee. #badweather

Experimentation

The prompt can be experimental and varied, without being limited to one format or style. ChatGPT can adapt to the prompt and generate different and surprising responses.

For example, if you want ChatGPT to generate a short story about a superhero who has the power to control time but uses this power irresponsibly, you might use a prompt like this:

Write a short story about a superhero who has the power to control time, but uses this power irresponsibly.

ChatGPT might respond with something like:

His name was Chronos, and he was the world's most powerful superhero. He could accelerate, decelerate, stop, or rewind time as he wished. He could save lives, prevent disasters, correct mistakes, and do whatever he wanted without worrying about the consequences.

But he didn't use his power for good. He used it for fun, revenge, advantage, and benefit. He didn't care about the side effects, reality changes, or temporal paradoxes. He only cared about himself.

One day, he decided he wanted to witness the origin of the universe. He traveled back in time, further and further, until he reached the moment of the Big Bang. He was fascinated by the cosmic explosion and decided to interfere. He tried to stop time to observe the phenomenon better.

But he made a grave mistake. He didn't realize that time was an inseparable dimension of space, matter, and energy. By trying to stop time, he halted everything that existed. He created a singularity point, where all physical laws ceased to apply.

He was trapped in that point, unable to move, think, or feel. He was condemned to an eternity of loneliness and suffering. He regretted everything he had done, but it was too late. He couldn't turn back.

He was the most powerful superhero in the world, but also the greatest villain. He was responsible for the end of the universe, and for his own hell.

Tone, Style, and Length in Prompt Engineering

When interacting with an artificial intelligence-based language model, it’s not just the words chosen that shape the response received, but also the tone, style, and length included implicitly or explicitly in the prompt.

Each of these elements plays a crucial role in directing the nature of the interaction, influencing everything from the formality of the response to the depth of the analysis provided. Exploring how these aspects influence the AI’s response can significantly enhance the efficacy of prompt engineering.

Tone

The tone used or requested in a prompt refers to its quality or expressed feeling, ranging from formal to informal, from serious to playful. The selected tone should reflect the desired type of response and the target audience. Some examples are:

• Formal: Used in professional or academic contexts, where precision and seriousness are paramount.
• Informal: More suited for casual conversations or when wanting to create a sense of closeness with the user.
• Professional: Marked by formal and direct communication, focusing on clarity and precision. Suitable for business contexts, official correspondence, and academic discussions where seriousness and respect are paramount.
• Friendly: Characterized by a warm and welcoming approach, often using casual language and expressions that promote closeness. Ideal for informal conversations, customer support, and any situation where creating a personal connection is desired.
• Enthusiastic: Conveys energy and positivity, showing excitement about the topic or situation. Effective for motivating the receiver, sharing good news, or simply enlivening the presented content.
• Empathetic: Demonstrates understanding and sensitivity towards the receiver’s feelings or situations. Essential in contexts of emotional support, conflict resolution, and whenever it’s important to show care and consideration.
• Instructional: Used to provide clear guidance or instructions. Objective and direct, ideal for tutorials, user manuals, and any educational content that requires clear steps to be followed.
• Reassuring: Aims to calm and offer security to the receiver, especially in situations of uncertainty or anxiety. Useful for mitigating concerns, providing support in crises, or when wanting to reinforce confidence in information or a plan.
• Inspirational: Employed to motivate and encourage the receiver, often appealing to their aspirations and desires. Effective for motivational speeches, personal development content, and situations where boosting positive action is desired.

Practical Example 1: Requesting an explanation of a quantum physics concept.

• Formal: “Could you elucidate the fundamental principles of quantum physics, emphasizing its theoretical and experimental implications?”
• Informal: “Can you explain to me how this whole quantum physics thing works? Like, the basics to start understanding?”

Practical Example 2: Explicitly stating the tone for the response.

• Formal: “In a formal tone, please explain the economic consequences of global climate change for emerging economies.”
• Informal: “In an informal way, tell me how artificial intelligence is changing the way we shop online.”
• Empathetic: In an empathetic tone, describe how someone might feel when learning a new skill and offer words of encouragement to overcome initial difficulties.
• Instructional: Use an instructional tone to guide someone through the process of preparing a healthy and balanced breakfast.
• Inspirational: In an inspiring tone, motivate someone to start a self-care journey, highlighting the benefits of dedicating time to oneself.

Style

The style concerns the choice of language and how we want the response to be structured, which can be narrative, expository, descriptive, persuasive, and many others. Some examples are:

• Narrative: This style is characterized by storytelling, focusing on characters, settings, and plots. The goal is to engage the reader through a cohesive narrative, often incorporating elements like dialogue, detailed descriptions, and character development. It’s widely used in literature, journalism, and even academic contexts to illustrate viewpoints or explore hypothetical scenarios.
• Expository: Focused on explaining or informing about a specific topic, the expository style aims to present facts, statistics, and arguments in a clear and logical manner. Unlike the narrative, which is built around a story, the expository is more direct, seeking to educate or elucidate the reader about a subject. It’s common in educational texts, scientific articles, and practical guides.
• Descriptive: The descriptive style aims to create a vivid image in the reader’s mind, using words to detail scenarios, objects, people, or feelings. The emphasis is on painting a picture through the rich and sensory use of language, allowing the reader to “see” what’s being described. This style is often employed in literary works, poetry, and marketing pieces that seek to evoke an emotional response.
• Persuasive: This style aims to convince the reader to accept a specific viewpoint or to take a desired action. It uses logical arguments, emotional appeals, and evidence, such as data or quotations, to influence the reader’s opinion or behavior. The persuasive style is key in speeches, advertising, business proposals, and opinion essays, where the goal is to change the audience’s perception or attitude.
• Analytical: This style focuses on breaking down complex information or concepts into smaller parts for detailed analysis. It’s common in literary analysis, movie critiques, and scientific research.
• Critical: Used to evaluate works, ideas, or arguments, highlighting strengths and weaknesses. Critical writing is essential in reviews and critical essays.
• Informative: Similar to expository, but with an emphasis on providing clear and factual information about a topic, without necessarily explaining or arguing. It’s common in instruction manuals and reports.
• Reflective: This style involves personal exploration of experiences, ideas, or events, often used in journals, blogs, and reflective essays.

Practical Example: Asking ChatGPT to write about the importance of environmental sustainability.

• Narrative: “Tell the story of a city that completely transformed its approach to environmental sustainability and the positive changes that occurred.”
• Expository: “Explain the principles of environmental sustainability and their importance for the future of the planet.”

Length

The expected length or size of the output can be requested in the prompt, ranging from a single word or phrase to a long and very detailed text, affecting the depth and breadth of the response.

• Short: Can be a single question or request, yielding direct responses.
• Long: Provides detailed context or multiple queries, leading to more elaborate and informative responses.

Practical Example: Asking for an analysis of current trends in artificial intelligence technology.

• Short: “What are the main trends in AI today?”
• Long: “Considering the rapid technological development and the increasing integration of artificial intelligence across various sectors, I would like a detailed analysis of the current trends in AI, including emerging applications and ethical implications.”

The length can also be determined by paragraphs, words, or tokens, each offering different degrees of control over the volume of the response.

Length Using Paragraphs and Words

Specifying the number of paragraphs or the word count are direct methods to define the desired length of an answer. These specifications help ensure that the response meets expectations for detail and length, whether for a concise summary or for an in-depth exploration of a topic.

• Paragraph Example: “Write a summary about the history of artificial intelligence in 3 paragraphs.”
• Word Count Example: “Develop a 1000-word article on the impacts of climate change on global biodiversity.”

Length Using Tokens

Tokens are basic units of text processed by language models like ChatGPT. A token can be a word, part of a word, or even punctuation. Specifying the number of tokens allows for more granular control over the response length, useful for adjusting the output extent in scenarios where the exact number of words or paragraphs is less important than the information load.

Generally, ChatGPT might treat a word as a token, but longer words or technical terms can be split into multiple tokens. For example, “ChatGPT” might be one token, while “artificial intelligence” might be counted as two tokens (“artificial” and “intelligence”).

• Numerical Example with Tokens: “Provide an explanation of the theory of relativity using exactly 50 tokens.”

When requesting responses from ChatGPT, incorporating numerical specifications about the desired length — whether in paragraphs, words, or tokens — not only helps to shape the depth of the topic exploration but also ensures that the responses better align with the available space or attention requirements.

By properly understanding and manipulating the tone, style, and length of prompts, it is possible to guide ChatGPT to produce responses that not only meet informational needs but also align with the desired approach and target audience. This flexibility in prompt engineering opens up a range of possibilities for creatively and effectively exploring topics.

Roles and Functions Using the “Act As” Technique

The “Act As” technique is a valuable method for directing the responses of a language model. It involves requesting the model to simulate the perspective or behavior of a specific entity, like an expert in a certain field.

By applying this technique, the user instructs the model to adopt the voice, knowledge set, and stance characteristic of the figure or role mentioned. For example, we might say: “Act as a marketing expert.” This shapes the response according to the expertise, communication style, and viewpoint expected of that persona.

Using the “Act As” technique in AI prompts is relatively simple but can produce highly personalized and relevant results. We have an article here on the blog, focused exclusively on this technique and in great detail: Master the “Act As An Expert” Technique in AI Prompts. In summary, it works like this:

1. Identify the Role or Persona: Start by clearly defining the role or figure you want the model to assume. It can be a finance expert, a math teacher, or even a poet.
2. Formulate the Appropriate Prompt: Create a specific prompt that directs the model to act as that persona. For example: “Act as a chef and explain the basic principles of French cooking.”
3. Enjoy the Personalized Response: Upon receiving the response, you will see that it reflects the perspective and knowledge of the chosen role. This allows for obtaining information or advice aligned with the typical experience of that figure.

Formulas and Structures of Prompts

In the art of prompt engineering, especially when interacting with advanced language models like ChatGPT, clarity, precision, and intentionality in formulating prompts are crucial for obtaining useful and relevant responses. A powerful way to enhance the quality of interactions with these AI models is through the use of prompt formulas or structures. These structures act as skeletons or molds that guide the construction of prompts, ensuring that all necessary elements for effective communication are present and well-organized.

Prompt formulas and structures help systematize the prompt engineering process, making it more accessible, especially for those taking their first steps in this field. They also serve as valuable tools for experienced users, offering a way to refine and experiment with different communication approaches. By employing these structures, we can significantly improve the accuracy of ChatGPT’s responses, as well as explore its capabilities for understanding and generating content more deeply.

In this section, we highlight four fundamental prompt engineering strategies: RTF, CTF, PECRA, and RISE. These structures were chosen for their proven effectiveness and the ability to provide distinct and efficient approaches to creating prompts, ranging from defining precise roles, tasks, and formats to detailed elaboration of contexts, objectives, and necessary actions. They represent just the beginning of a vast array of possibilities for enhancing your interactions with ChatGPT, paving the way for clearer, more objective, and fruitful communication.

For those interested in exploring the diversity and depth of available strategies in prompt engineering further, we invite you to read our exclusive article dedicated entirely to prompt formulas and structures. In it, we dive into an additional variety of methods, each offering unique and valuable perspectives for enhancing your requests and obtaining even more relevant and personalized responses.

RTF: Role, Task, Format

The RTF (Role, Task, Format) structure is an effective approach to creating clear and directed prompts, establishing a framework that defines the requester’s role, the specific task to be performed, and the desired format for the response. This structure helps guide the language model, like ChatGPT, to produce outputs more aligned with the user’s expectations, both in terms of content and presentation.

Practical example of RTF:

As a finance expert (Role), analyze the current trends in the stock market (Task) and present your findings in a list of key points (Format).

ChatGPT might respond with something like:

1. Market Volatility: The stock market has shown significant volatility, influenced by global economic factors and political uncertainties.

2. Technology on the Rise: Shares of technology companies continue to demonstrate strong growth, driven by digital acceleration.

3. Sustainability as Focus: There is a growing trend of investing in companies with sustainable practices, reflecting a shift in investor values.

This example demonstrates how the RTF structure can be employed to obtain a specific analysis from ChatGPT, directing the model to focus on the role of a finance expert, perform a market analysis task, and format the response in an accessible and informative manner.

CTF: Context, Task, Format

The CTF (Context, Task, Format) structure is a powerful tool in prompt engineering, allowing users to clearly establish the scenario in which the interaction occurs, the specific action expected from ChatGPT, and how the response should be structured. This approach facilitates the creation of prompts that are both precise in their requests and specific about the desired outcome, optimizing the effectiveness of the response obtained.

Practical example of CTF:

Given the current context of growing interest in sustainability practices (Context), develop a practical guide for companies wishing to implement environmentally sustainable measures in their operations (Task), organizing the information into topics with subtitles for each suggested measure (Format).

ChatGPT might respond with something like:

• Introduction to Corporate Sustainability: A brief overview of the importance of adopting sustainable practices.

• Waste Reduction: Strategies to minimize waste production, including recycling and composting.

• Energy Efficiency: Measures to improve energy use, such as installing renewable energy systems and upgrading to low-consumption equipment.

• Sustainable Supply Chain: Tips for choosing suppliers that adopt environmentally responsible practices.

This example illustrates how the CTF structure can be used to request a detailed and well-organized guide on a specific topic, with ChatGPT generating content that not only fulfills the designated task but also follows the required format, facilitating the understanding and application of the information.

PECRA: Purpose, Expectation, Context, Request, Action

The PECRA (Purpose, Expectation, Context, Request, Action) structure is a detailed tool for crafting prompts that guide ChatGPT to deliver responses that not only meet a specific request but are also aligned with the user’s purpose and expectations. This method is particularly useful for ensuring that all nuances of a request are addressed, offering a clear path to more accurate and informative responses.

Practical example of PECRA:

Considering the growing interest in sustainable living habits (Context), create an introductory guide (Request) aimed at individuals who wish to start adopting more sustainable practices in their daily lives (Purpose). The guide should offer practical and accessible suggestions (Expectation), organized into sections addressing different aspects of everyday life, such as food, transportation, and consumption (Action).

ChatGPT might respond with something like:

The introductory guide would start with an overview of the importance of sustainability, followed by sections dedicated to:

1. Sustainable Eating: Tips for choosing local and seasonal products, and how to reduce food waste.
2. Green Transportation: More sustainable alternatives to using private cars, such as bicycles, public transport, and carpooling.
3. Conscious Consumption: Strategies to reduce the use of single-use products and opt for reusable or recyclable options.

This example illustrates how the PECRA structure can be used to formulate a complex prompt, guiding ChatGPT in creating content that not only meets a specific information need but is also structured in a way to reflect a clear purpose and meet defined expectations.

RISE: Role, Input, Steps, Expectation

The RISE (Role, Input, Steps, Expectation) structure is a methodology designed to guide prompt formulation in a way that clearly outlines the requester’s role, the information or inputs needed for the task, the specific steps for completing the task, and the expectations for the outcome. This structured approach is ideal for situations requiring detailed instructions or for processes involving multiple steps.

Practical example of RISE:

As a health and wellness consultant (Role), using the latest information on nutrition and physical exercise (Input), develop a 4-week plan to improve the lifestyle of someone starting to exercise (Task). The plan should include weekly dietary recommendations and a progressive exercise schedule (Steps), aiming to increase energy, improve sleep, and reduce stress (Expectation).

ChatGPT might respond with something like:

A detailed plan starting with an introduction to basic principles of nutrition and exercise, followed by:

1. Week 1: Focus on hydration and low-impact exercises to build habit.
2. Week 2: Introduction of a balanced diet emphasizing vegetables and lean proteins, combined with moderate exercises.
3. Week 3: Increase in exercise intensity and introduction of mindfulness techniques for stress reduction.
4. Week 4: Evaluation of progress, diet adjustments as needed, and an increase in exercise intensity to promote continuous improvements in energy, sleep, and stress management.

This example shows how the RISE structure can be applied to create a detailed and action-oriented prompt, enabling ChatGPT to develop a lifestyle improvement plan that is both informative and practical, aligned with the provided inputs and established expectations.

Reminder before we proceed: we have an exclusive article on prompt formulas and structures that covers more than these 4 examples, including: TREF, CREATE, CREO, PAIN, GRADE, ROSES, RDIREC, RSCET, RASCEF, APE, TAG, ERA, RACE, COAST, and SPARK.

Intermediate and Advanced Prompt Engineering Techniques

In prompt engineering, especially when working with advanced language models like ChatGPT, the adoption of advanced techniques can significantly enrich the quality and accuracy of the responses obtained. These techniques, which go beyond basic question-formulation approaches, allow for more sophisticated and tailored interaction, capable of deeply exploring the model’s capabilities and generating outputs more aligned with the user’s specific needs.

Next, we will explore some of these advanced techniques, including prompt patterns, thought chains, nested prompts, the use of metadata, as well as rephrasing and feedback prompts, offering practical examples to illustrate how each can be applied.

Reusable Prompt Templates

Reusable prompt templates refer to the strategic use of repetitive formats or structures that guide the model to respond in a specific manner. These patterns can be especially useful for maintaining consistency in responses to a series of related questions or for training the model to recognize a desired response format.

Practical Example: Asking ChatGPT to create titles for a series of blog posts about technology.

• Prompt Template: “Create an attractive title for a blog post about [topic]: [specific topic].”
• Application:
  • Topic 1: “Create an attractive title for a blog post about technology: AI Trends for 2024.”
  • Topic 2: “Create an attractive title for a blog post about technology: The Future of Autonomous Vehicles.”

Thought Chains

Thought chains are a technique where the prompt encourages the model to “think out loud,” detailing the logical process or reasoning steps leading to a conclusion. This not only provides insights into how the model reaches its answers but also helps create more detailed and substantiated responses.

Practical Example: Explore the impact of machine learning in medicine.

• Prompt: “Explain, step by step, how machine learning is transforming medical diagnosis.”

Nested Prompts

Nested prompts involve embedding multiple requests or questions within a single prompt, allowing the model to address a sequence of related topics in a structured manner. This can be particularly useful for deeply exploring a complex theme.

Practical Example: Investigate the relationship between diet and mental health.

• Prompt: “First, describe how a balanced diet influences mental well-being. Then, provide examples of foods that promote good mental health.”

Metadata

The use of metadata in prompts refers to the inclusion of additional or contextual information that helps the model better understand the nature of the request or the desired response format. This can range from specifying the tone or style to setting criteria for structuring the information.

Practical Example: Request a summary of a scientific article.

• Prompt with Metadata: “In an academic tone, summarize the scientific article titled ‘Effects of Global Warming on Marine Ecosystems’, focusing on the main findings and conclusions.”

Rephrasing and Feedback Prompts

Rephrasing and feedback prompts are used to refine or adjust previous responses, asking the model to consider criticisms or suggestions and apply them to improve the response. This technique encourages an iterative process of enhancement.

Practical Example: Refine an explanation of blockchain.

• Initial Prompt: “Explain what blockchain is and how it works.”
• Feedback: “Your explanation was good, but can you make it more accessible to laypeople? Focus on simple analogies to explain blockchain technology.”

Ask Before Answering

The “Ask Before Answering” technique is a prompt engineering strategy that consists of requesting ChatGPT to ask specific questions before providing an answer. This aims to clarify the user’s intent, ensure all necessary information is present, and thus produce more accurate and aligned responses.

Practical Example: Suppose you want detailed information about the implementation of neural networks in artificial intelligence projects. A direct prompt could be: “Explain how to implement neural networks in AI projects.” However, using the “Ask Before Answering” technique, the prompt transforms into an instruction for ChatGPT to formulate clarifying questions before offering the explanation:

• Initial Prompt: “Before explaining the implementation of neural networks in AI projects, what specific information do you need to know to provide a detailed and accurate answer?”
• ChatGPT’s Questions:
  • “Are you looking for information about the architecture of neural networks specific to some type of application?”
  • “Is there a preferred development framework for the implementation?”
  • “What is the target audience’s level of experience with artificial intelligence and neural networks?”
• Response Based on Questions: Based on the answers to these questions, ChatGPT can then provide a more targeted and useful explanation, covering specific aspects of neural network implementation, recommending suitable frameworks, and adjusting the complexity of the explanation according to the audience’s level of experience.

This approach promotes a more dynamic and efficient interaction, allowing the model to adjust its response based on a deeper understanding of the user’s needs. Moreover, it helps avoid misunderstandings and optimizes time by providing information that is truly relevant and valuable to the requester.

These prompt engineering techniques open new possibilities for interactions with ChatGPT, allowing for more complex requests and more refined responses. By mastering these approaches, users can maximize the potential of language models for a wide range of applications.

Combining Elements in Prompt Engineering

In the art of prompt engineering, the key to creating rich and effective interactions with ChatGPT lies in the strategic combination of different elements such as principles, structures, tones, styles, lengths, and advanced techniques. By integrating these components, it’s possible to shape prompts that not only communicate the request clearly but also guide the model to generate responses that align precisely with the specific needs, context, and goals of the user.

Integrating Elements for Effective Prompts

Each element brings its unique contribution to the formulation of the prompt:

• Principles establish the foundation for clear and objective interactions.
• Structures provide a skeleton to organize the request.
• Tones adjust the emotional or formal approach of the response.
• Styles direct how the information is presented.
• Lengths define the detailed extension of the expected response.
• Advanced Techniques allow for deeper explorations or specific response formats.

By deliberately combining these elements, users can create prompts that not only specify what they want to know or do but also how they want the information to be delivered, reflecting considerations such as depth of analysis, emotional tone or formality, and suitability for the target audience or medium of communication.

Practical Example 1: Requesting an analysis of the stock market

• Principle: Clarity and specificity
• Structure: PECRA
• Tone: Professional
• Style: Analytical
• Length: “Write a 500-word summary”
• Technique: Metadata to specify topics of interest

Combined Prompt: “In a professional tone and using an analytical approach, provide a detailed analysis of the current stock market, focusing on the technology and health sectors. Use the PECRA structure to organize your summary, which should be approximately 500 words, including metadata on price trends, trading volume, and future projections.”

Practical Example 2: Creating a guide for gardening beginners

• Principle: Step-by-step instruction
• Structure: RISE
• Tone: Friendly
• Style: Instructional
• Length: “Develop a guide in 3 paragraphs”
• Technique: Thought chains to explain the “why” behind each step

Combined Prompt: “With a friendly tone and an instructional style, create a step-by-step guide for gardening beginners, using the RISE structure. The guide should consist of 3 paragraphs, covering plant selection, soil preparation, and the routine of basic care, including thought chains to explain the importance of each step.”

By coherently integrating these elements, it is possible to create prompts that guide ChatGPT to produce highly relevant, informative, and engaging responses, maximizing the model’s potential to meet a wide variety of requests in a customized manner.

Supplementary Articles

This article was immense and brought a significant amount of relevant information, but believe it, it’s just the tip of the iceberg. I am planning various other complementary and related articles, which I will list below, and will be including as they are published:

• 19 Formulas and Structures of Prompts for ChatGPT: Going Beyond the Basics
• Exploring ChatGPT Output Formats: A Guide to Presenting Information
• Master the “Act As An Expert” Technique in AI Prompts
• Using Copywriting Techniques in AI Prompts

Frequently Asked Questions

Conclusion

Prompt engineering is an essential skill to leverage the full potential of ChatGPT and other large language models (LLMs), which can have diverse and disruptive applications in various fields and sectors. Prompt engineering involves creating natural language instructions or queries that guide ChatGPT to generate the desired output, using principles, formulas, and structures, techniques, and elements that can improve the quality and relevance of the model’s responses.

Prompt engineering is not just a technical skill; it is a bridge between human creativity and the computational power of artificial intelligence. We understand that by mastering the art of formulating effective prompts, we are able to communicate more deeply and meaningfully with advanced language models, thereby unlocking new horizons of creativity, efficiency, and discovery.