8 Easy Tips to Craft the Perfect GPT Prompt Every Time
- David Wills
- Jul 25, 2023
- 7 min read
ChatGPT completely changed the writing landscape. Whereas before we had to write everything from scratch, we now have an incredible tool that can either provide a solid first draft of pre-written ideas or, in many cases, even produce the exact piece of writing we want with hardly any changes. Articles that used to take hours can now be put together in a matter of minutes and with a fraction of the effort.
But like any powerful tool, its effectiveness lies in how skillfully we wield it. Crafting the perfect GPT prompt isn’t just about getting a response - it’s about shaping that response to fit your precise needs.
I've worked extensively with AI text generators for months now and have discovered a few techniques that have helped me produce prompts that get the best results pretty much every time. Here are some of my personal tips for crafting the perfect GPT prompt every time.
1) Start With A Clear Idea Of What You Want To Produce
You must first begin with a clear goal of what you want to achieve, as this will help you determine the instructions you give the AI. This primarily includes content and structure but also extends to style, voice, and tone.
For instance, let’s say we want to write a blog post about learning to meditate. We could easily just prompt the AI with something basic like:
“Write an article about how to meditate.”
The AI will write an article, but it’s going to be very generic, and it may not even give it to you in the format that you want.
But if we have a specific structure, tone, and voice that we’re aiming for, we can get much more dynamic results, such as with this prompt:
“You are a blog writer. Write a six-item blog post detailing how to learn to meditate, including different meditation techniques and tips on how long to meditate for. Start with two introductory paragraphs that give context to the benefits of meditations. Then for each item, write the item as a sub-header. Then write two paragraphs explaining the item. Use a casual but detailed style. Use vivid descriptions with hints of humor and sarcasm. Use a confident, authoritative voice that maintains a sense of enthusiasm and excitement for the subject matter.”
Notice how granular we can get with our descriptions. The more we can hone in on these meta concepts - such as structure, tone, and voice - in our prompts, the better our end results will be.
2) Tell It What Role You Want It To Play
Notice in the previous example that I told the AI to be a blog writer. By telling the AI what role you want it to play, you prime its response to generate as if it were in that role.
For instance, if I were a manager writing a first pass of performance reviews, I might say something like:
“Write a performance review for Jeff that highlights his sales copywriting skills over the previous year.”
This is fine, but it doesn’t give the AI any real context of who we are in relation to Jeff or what key points to focus on. Instead, we should instead say something like this:
“You are a manager in the marketing department at a software company. Write a performance review for your direct report, Jeff, that highlights his sales copywriting skills over the previous year while citing his room for improvement in speed and efficiency.
Now the AI will use the context of the type of company (“software company”) as well as the finer points that we want to address in this performance review (“room for improvement in speed and efficiency”). Of course, this is just a basic example, and the more granular you can get with specific details you want it to touch on in its response, the less likely you’ll need to give follow-up results
3) Give Detailed Instructions For Structure And Style
One of GPT’s greatest strengths is generating text in hyperspecific structures and styles. You’ve likely heard of prompts asking for things like zombie love poems written in the style of Shakespeare, and the mere fact that GPT can generate something like this speaks to the power and range GPT has when given parameters of structure and style.
This is truly where the magic of GPT occurs and where good responses turn into excellent responses. By giving detailed responses for the structure and style, you can produce radically different results under the same general topic. This is why it is so crucial to have a clear idea of what you want to produce from the get-go (see item #1).
For the computer science-minded folks out there, you can also define structure in terms of variables. For instance, if I wanted to reformat a list of movies, I might prompt something like this:
“You are an editor. I am going to give you a list of movies, and you are going to reformat them in the structure of “[MOVIE] ([YEAR])” where [MOVIE] = the name of the movie and [YEAR] = the year that movie was released.”
This sample really only scratches the surface of what is possible with this line of thinking. The more you can think like the AI the better, and at the end of the day it really is just a sophisticated computer.
4) Use The AI To Analyze Similar Writing Examples
If you’re quite not sure how to describe the style you’re looking for, one of my favorite GPT tricks is to feed it text I’ve already written and have it analyze it for tone, style, language, and voice. This can prove incredibly effective when you have an example of writing you want replicated but aren’t quite sure how to describe that piece of writing.
A common prompt I use is:
“You are a sentiment analyzer. I will provide you with text and you will analyze it for style, tone, language, and voice. Analyze any potential rhetorical techniques being used as well.”
What you will generally get in return is a detailed breakdown of how a piece of content was written, which will help you hone in on the specific keywords to include in your prompt. This will help you put the more abstract qualities of writing into language that your AI tool can understand.
5) Cut Out Unnecessary Fluff
While it’s important to have detailed responses, you don’t want your prompts to contain unnecessarily verbose language. The more concise you can keep your prompts, the less opportunity there will be for the AI to misinterpret your actual goals.
For example, we could prompt our AI by saying something like:
“I would kindly entreat you to dedicate your time and intellectual efforts towards the crafting of a comprehensive and extensive digital article, more commonly referred to as a blog post, that delves into the intricacies of acquiring the skill set and understanding requisite for the ancient and introspective practice of meditation. Additionally, it would be of paramount importance to ensure the inclusion of detailed expositions on the myriad methodologies and techniques associated with said practice. Furthermore, do not neglect to underscore and elucidate the numerous advantages and beneficial outcomes that one might expect to reap from the diligent and consistent engagement in meditation on a daily basis.”
This is a cartoonishly long example way of saying:“Write a blog post about learning how to meditate. Be sure to include details about different practices as well as the benefits of practicing meditation every day.”
Of course, it’s very likely nobody will ever write out the first prompt, but it illustrates a style of writing that does not work well with AI. Keep your prompts as short, sweet, and direct as possible. Don’t be afraid to tell it exactly what you want it to do (it’s a program that’s designed to do exactly what you tell it to do, after all, so don’t be shy).
6) Use Examples In Your Prompts
We can also feed the AI examples of our desired output within the prompt itself. Providing explicit examples of our desired goal can dramatically improve the quality and accuracy of our results, as we are giving the AI a model to work off of. This is called few-shot prompting (deemed so because you’re offering “a few examples”), and it’s an incredibly valuable technique for
Opting not to include examples would be like trying to train someone to draw an apple if they’ve never seen an apple: sure you can describe it in broad strokes and hope for the best, but the finished product would be much better if they were to actually see an apple first. The same principle applies here.
Generally, you want to be explicitly clear in your prompt which text is part of an example and which text is part of a prompt. This can be done in multiple ways, but I tend to simply put each one in quotation marks and label it as “Example 1”, “Example 2”, etc.
7) Iterate And Refine Using Follow-Up Prompts
The chances are high that the AI’s first pass won’t be what you hoped it would be. That’s why it’s crucial to iterate and refine with follow-up prompts.
Pay close attention to the differences between your intended result and your actual result. What elements of the writing aren’t quite how you wanted it to be? Is the structure off? Do you want it to have a certain tone? Is it including irrelevant content?
You can then ask follow-up prompts to refine its original result. You could also just start a new conversation, which may be useful if you’re iterating in order to create a reusable prompt, but very often it makes more sense to continue in your current GPT chat so that it remembers its previous answers.
8) Save Prompts For Future Use
There’s truly no need to reinvent the wheel every time. In fact, many of your prompts may follow the same basic skeleton, with minor variations in the finer details of content or style. I recommend maintaining a spreadsheet somewhere with successful prompts that you can reuse in the future.
Depending on how granular you decide to get, you can even break this spreadsheet down with different columns/rows for role, structure, content, style, tone, etc. However, even just storing successful prompts someplace (that isn’t the chat history) can go a long way to toward ensuring you are always getting the most effective prompts possible.
Conclusion
The more you interact with the AI, the more you’ll become accustomed to, for lack of a better term, how it thinks. Ultimately these AI tools are just sophisticated tools, and the more we use them, the better we’ll understand how to adapt to the unique ways they operate.
Have you used any of these tips in your prompts? Are there any additional tips that you’ve found useful in generating excellent text from AI text generators? Let me know in the comments. Thanks for reading, and I’ll see you in the next post!
Comments