Disclaimers
Affiliate Disclaimer:This site contains affiliate links. If you make a purchase through these links, we may earn a commission at no extra cost to you. We only promote products we believe will add value to our readers.
For more information, please see our Affiliate Disclaimer Page.
Financial Disclaimer:Investing and trading in stocks, bonds, and other financial instruments carries inherent risks. The content on this blog is not intended to serve as financial or investment advice. Please consult with a certified financial advisor before making any investment decisions.
While we strive to provide accurate and up-to-date information, we do not guarantee the accuracy, completeness, or timeliness of any information provided. Any reliance you place on the content of this blog is strictly at your own risk.
Please see our Disclaimer Page.
This content is free thanks to our sponsors
Sponsored Ad – I may earn a commission if you click and buy
Table of Contents
What prompt engineering is
Final: Prompt engineering is the ability to formulate queries to an artificial intelligence so that the results are maximally useful for the user. Simply put, a prompt is a message sent to an AI, for example, a request or a question. However, engineering is a process of developing a prompt: choosing words, phrases, and formulations so that the response meets all the requirements and expectations. In short, it is about learning how to ask the right questions.
If that sounds intimidating, here's some good news. You already know most of it! Imagine asking a new talented assistant to help you, someone clever but who has never met you and knows nothing of your circumstances. If you said, write something about dogs, you'd probably get something generic. If you explain who it's for, what it should do, and how long it should be, you'd get much better results. Prompt engineering is simply communicating clearly, applied to a machine.
Why is this important? Modern AI chatbots are based on large languange models, which predict what text would be most useful given the request. They are impressive, but can only fulfill requests they understand. They cannot read minds. Only you know exactly what you want, and vague requests can result in poor results. The more specific you are in your request, the better a job the model will be able to do.
The building blocks of a good prompt
The first element that is needed in a prompt is a simple task. Ask the AI to do something straightforward and specific. For example, instead of asking to help with my email, try rewriting this email so that it sounds friendly but still professional while being concise and under one hundred and fifty words. Using a task-based prompt usually means the response will be more organized and straightforward. Use action words like summarize, compare, explain, draft, translate, or list in the request. This lets the AI know exactly what to do.
The second ingredient is the context. Tell the artificial intelligence what background information it needs to know to respond appropriately to the user's question or request. Who is the request aimed at? What circumstances does it occur in? What has the user already done? For example, you may ask to write an email to a customer who has been waiting for three weeks for a refund, apologizing and explaining the situation. These details will have a tremendous impact on the phrasing of the response. It is often the context that makes the difference between a bland and an unhelpful answer.
Third, you may want to assign the AI a role. You might ask it to act as a patient math tutor for a twelve-year-old, or you are an experienced editor who provides gentle, practical feedback. A role helps establish the voice, level of detail and perspective of your content. It won’t make the AI an expert, and you should always fact check, but it can guide the style in a helpful direction.
Fourth, describe the format you would like. Do you want a short paragraph, a table, a numbered list of steps, an email, a script? Do you want it in layman's terms or technical speak? You want it to be warm and friendly or stiff and formal? If you leave this out, the AI may decide for you. It may not be what you want. Stating what you want it to look like will save you time and headache.
Fifthly, it is essential to specify any constraints or limitations. It may relate to the word count, certain prohibited or permitted information, requirements to the writing level, or specific points that should be covered. For instance, the task may impose a necessity to explain everything to a novice easily, exclude complex terms, and complete the assignment within three short paragraphs. Overall, constraints are a limitation of the means or resources, and, similar to the small canvas in the analogy above, they challenge the author to create a remarkable result.
Examples, steps and iteration
One of the most effective methods is asking to provide an example. If you give an example of what you want, the AI can mimic that pattern. For instance, if you need product descriptions that follow a specific style, pasting two similar ones and asking for the third in the same vein may work. It is sometimes called few shot prompting since you are giving an example. Describing vague ideas can be challenging, while demonstrating is often more effective.
You can also ask the AI to think step by step, for example, for mathematical problems or planning or comparing options. You can invite the AI to think through things carefully and explain their reasoning before jumping to a solution, especially on challenging questions. Some newer systems do a decent amount of that thinking on their own, but it can help to prompt them for particularly thorny issues.
Iteration is a huge part of the craft. It's rare that something comes out perfect on the first pass, and that's normal. See what you got, and ask yourself what's missing. Was it too lengthy? Too stiff? Did it miss something? Then, iterate! Either by rewording the prompt or replying with directions as to what to change. Think of it as a conversation where you lead them to where you want to go.
It helps to be specific when giving feedback. Just saying that's not good is not going to help the AI much. If you say that's not good enough, the opening was stiff, and I want a more casual tone with shorter sentences, then the change is going to be worth it. It works the same way with people. The better you can frame what you want, the faster you're going to get a result that you can live with.
Here is a short example that demonstrates the difference. A prompt could be “give me a workout plan” and the improved version would be “I am a forty-year-old novice with no equipment and thirty minutes to spend daily on a home workout. Develop a four-week plan for me with a gradual intensity increase, organized as a weekly table, and a note to consult a physician if I have any health conditions.”. The second option would produce a more detailed and relevant answer.
Another habit that might be helpful is to separate your instructions from the material that you want worked on. So if you paste in a long article and ask for a summary, it might help to put brackets around the article or otherwise indicate where the article starts and ends. This will help the AI distinguish your instruction from the text you actually want it to work on and will avoid issues if the text you're asking it to work on contains instructions or questions.
For longer or more involved tasks, break it into pieces. Instead of saying, “Write me a business plan,” ask for an outline first, and then for the various sections. This is sometimes called prompt chaining. It can be very helpful in keeping control of what the model does. It enables you to review and correct each part before proceeding. It's like writing a book chapter by chapter rather than trying to write the whole thing at once.
Many AI tools allow you to set instructions that will apply to all of your interactions with the system. These are sometimes known as the “system prompt” or “custom instructions.”. The instructions could be something as simple as “be polite” or as detailed as “you are a first-year associate at McKinsey & Company.” System prompts enable developers to customize the behavior of their AI chatbot. However, for average users, these prompts serve to remind the AI of specific behaviors without having to type them in every time.
Accuracy, privacy and tool differences
It is always a good idea to remember that the AI can confidently give wrong answers. Sometimes, it can give detailed and plausible, but essentially false information, so one should always be careful. This phenomenon is called hallucination, and prompt engineering can help alleviate it, for instance, by asking the model to say when it is unsure of something or using only the text provided. Nevertheless, it cannot be defeated completely, so double-checking any important information from a credible source is advised.
You can reduce errors by asking the AI to show its sources or its reasoning, and then verifying this information. Beware that some AI systems may make up citations that don't actually exist, and if one is important to you, double check it. A good prompt should tell the AI, if it doesn't know an answer, to say as much, and not guess. You should give it permission to be honest if it doesn't know something.
Another consideration is privacy. Avoid putting the passwords to your accounts, financial information, confidential work documents, or the private information of others into an AI tool, unless you are certain how the information will be used. Different services may have different policies, and different levels of security. As a rule of thumb, only share information which you would be comfortable having a third party see. If in doubt, leave out names and personal information, or replace them with placeholders.
Different artificial intelligence tools require different prompts. A prompt which works well for one tool might need to change slightly for another one. Some artificial intelligence tools work better with certain types of writing than others. It can be worth it investing time learning how different artificial intelligence tools work and what prompts help them produce the best results. Keeping a record of prompts that work well for different artificial intelligence tools can save hours of frustration.
Prompt engineering is of great importance, not only for casual communication but also for the creation of customer service bots, writing assistance, research, teaching tools, lesson planning, coding assistance, and coding documentation for programmers, among other things. In all these cases, it is necessary to create test prompts and continuously optimize them through iterations. This has turned into a full-time occupation for some people.
Professionals test their prompts, as they would any other work. They try their prompts on numerous samples, looking for flaws and comparing different approaches to see which works best. A good prompt should not only work for one case but be reliable for all scenarios. If you build a product for others, budget the time to test your work and find examples of what does and does not work.
Common mistakes
The most common mistakes people tend to make while using the prompts are easily avoidable. One of them is providing vague information. Another mistake is including too many different requests in one prompt, which might confuse an AI. Furthermore, many users forget to specify the target audience and format the response in the desired way. Finally, some people tend to provide opposite instructions such as making the text short and including all possible details. All these mistakes can be noticed on the first sight by a person who would read this request in a prompt.
It also helps to specify what one wants instead of what one does not want. The instruction 'don't be boring' is vague, and offers the AI plenty of latitude. On the other hand, an instruction such as 'use a friendly tone with some humor' is a more precise guide to the desired outcome. Whereas negative instructions merely specify what is not desired, positive instructions offer the AI a goal to work towards. There is no doubt that a combination of both types of instructions will yield desirable results, but it is important to give preference to positive instructions.
Tone and politeness are a minor curiosity. People often wonder if saying please and thank you makes any difference. It rarely makes much difference in terms of quality, but being clear and polite is a virtue, and many people feel it helps keep their own communications warm. What matters much more is the content of the request. Spend your effort on making it clear and contextualized rather than trying to think of clever tricks or magic phrases.
Speaking of tricks, there are many articles and guides online that offer some secret prompts that will supposedly allow one to obtain incredible results. However, all of them are just a fancy way of describing the obvious instructions. There is no secret, just a simple explanation of what one wants or needs, and a description of what they actually get. Obviously, if something works, it should be used, but not employing some additional fluff, suggesting that there is some magical formula that has to be entered in a certain way.
As AI evolves, some things seem less relevant, as new models are better at handling ambiguities and are more accurate with their responses. However, the fundamental principles still apply, and it is important to know one’s needs and specify them correctly in order to receive adequate results. Some professionals suggest talking about context engineering, which refers to engineering information that an AI would use as a basis for calculating answers, beyond the query itself.
A full example, and practice
Let's walk through a complete example, together. Let's say that you need help drafting a thank-you note to a teacher. First pass: write a thank you note. The AI responds with something generic. You refine your query: write a short, heartfelt thank-you note to my daughter's third grade teacher, Ms. Reyes, who helped her gain confidence in reading this year. Keep it under one hundred words, and avoid being overly effusive. Now you have a thank-you note worth sending.
Notice what changed. You added the audience, reason, length, and vibe you wanted. No hard technical specs. Just telling the story so it would get done right, and then the AI had more to work with. And that's what prompt engineering is all about: Not having to know all the rules, just having the information that an intelligent human assistant would need to do a good job.
If you wish to practice, you may attempt several simple experiments. Ask the same question in three ways: vaguely, with more detail, and with an example. Compare the answers. Ask the same information in different formats. Ask the system to critique the answers and see if they are improved. These experiments will teach you far more than any set of instructions, as you will see precisely how your language is affecting the results.
You can also ask the AI to help you to write better prompts. Ask something like, I would like to ask you to help me to plan a vegetable garden. What questions you should ask me first to help me? Often the AI will provide with an excellent list of questions that you did not even think about. It makes the process fun as it becomes like an interview, which is always a good start if you have no idea how to describe your needs.
Besides the obvious ethical concerns of using such technology, it is essential to keep in mind that not everything can be solved by an algorithm. There is value in the judgment and responsibility of a real human being. One must not use such programs to fool others, commit plagiarism, or create toxic content. When used in education or professional writing, one has to remember the specific standards of conduct and always remain transparent about the methods and materials used. Prompt engineering is a helpful tool, but it should not be abused, for the well-being of society as a whole.
With the passage of time, one will see a pattern that will help them know what works, what does not work, and what details are needed when giving instructions or what the preferred method of communication is. It is similar to learning how to conduct a search correctly. One types in nonsense at first and then learns the correct keywords that yield the best results. It does require time to learn, but the information helps one save time in the long run.
Length is always an issue. How long should the prompt be? Short or long? The truth is that it should be as long as it needs to be. If you have a simple task, a short prompt will do. But if you need to create a report based on a set of disorganized notes, you will need more detailed instructions and a more extended prompt. Don’t be afraid that your prompt will be too long; just make sure that your sentences are organized, and the essential parts are highlighted.
Creative work benefits from a dash of playful experimentation. When asking an AI to help you come up with a story, a poem, a catchy name for your new bakery, etc., be sure to provide some flavor text to go along with your request. You can specify the mood you're looking for, the words you want to include or avoid, give it something similar that you like, and ask for multiple options. Getting a few options to choose from can be more rewarding than asking for one specific result.
When making queries in the application of data or documents, make sure to specify what information the AI should use as its foundation. You could write, use the text below to answer, and let me know if the answer is not here. This way, the AI will rely on your information and not its general knowledge database, which is often more important for summaries of reports, questions about contracts, or even extracting specific dates of importance.
Finally, remember that you are in charge. While the AI is a helpful tool, it is up to you to judge what is good enough, what is accurate enough, and what is appropriate to use. You should read over the results with a critical eye, editing anything that seems wrong, and trusting your knowledge. After all, the best results come from a collaboration between your own knowledge and the AI's quick and vast database, and can be much better than either could do alone.
So, in conclusion, prompt engineering could be defined as the art of communication with the help of which one can obtain the desired results from the AI. The instructions should be precise and well-structured, and it is essential to edit and tune them until the moment the required outcome is achieved. In addition, one should state the request, provide the background and context, specify the format and rules, and give examples. It is also vital to check the information and data, stay private and safe, and, finally, experiment.
