The introduction of ChatGPT has marked a turning point in the evolution of AI-based language models. From drafting emails to writing compelling articles, this language model has become an indispensable asset to many.
However, it’s not without its critics. Some complain about its alleged poor writing skills or tendencies to plagiarize, while others worry about the quality of its output and its trustworthiness.
After spending 200+ hours with ChatGPT, I came to the conclusion that the tool itself is only as effective as the prompts it receives. The lack of skill in framing these prompts is the real culprit here, and even just learning the basics of prompt engineering can completely transform the quality of the results you get from ChatGPT.
What is prompt engineering?
Prompt engineering is the science and art of designing effective prompts to extract desired responses from an AI model like ChatGPT. Similar to asking insightful questions, the quality of your prompts will determine the caliber of answers you receive. This is critical as AI chatbots operate on the well-known ‘Garbage In, Garbage Out’ principle (GIGO), implying that your input directly affects the quality of the output.
Why is prompt engineering so crucial? Well, consider ChatGPT your diligent AI assistant who takes your instructions at face value.
If your prompts are vague, you’re likely to get misunderstood or subpar responses. It’s actually pretty similar to human assistants: failure to clearly communicate the exact requirements often results in mutual disappointment and unmet expectations.
On the contrary, clear and specific instructions are likely to produce significantly more accurate and relevant responses. Mastering this art of communication with AI unlocks the full potential of tools like ChatGPT, leading to productive and efficient interactions every professional could benefit from.
Common mistakes in prompting
To understand what makes a good ChatGPT prompt, let’s first look at the most common mistakes that often occur in the process of writing it:
- Vague instructions: Similar to communicating with humans, ambiguity with AI leads to misunderstandings. A lack of specific instructions often results in broad, generic, or unhelpful responses.
- Lack of context: ChatGPT doesn’t have a memory of past prompts or responses unless it’s within the same conversation, and even then, it’s limited, and at some point, it “forgets” the beginning. If the essential context is missing, the quality of the output can significantly decline.
- No styling instructions: Many users forget that they can ask ChatGPT to generate text in a specific style or tone. Without such instructions, the AI will default to a neutral style, which many find generic, boring, and unusable for publishing.
- Zero personality: By default, ChatGPT’s responses are quite formal and lack a personal touch. It doesn’t know your natural style or your brand voice – you have to teach it. Think of it as onboarding a new team member.
- Assuming that AI knows everything: ChatGPT has been trained on an impressive amount of data, but it doesn’t know everything. Sometimes you need to point to a specific concept or a source website as a starting point. It is also known for its hallucinations, so it’s important to fact-check everything it says and not use it as your only source of information.
- Forgetting to set constraints: If you want ChatGPT to avoid certain topics, styles, or even words, you should specify those constraints in your prompts. The AI can’t read your mind (yet!) and doesn’t know what is off-limits unless instructed.
Guidelines for effective prompting
Since you are now aware of the most common mistakes, here are the prompting guidelines you can use as a reference to perfect your communications with ChatGPT.
Always provide context
When you start a new conversation, ChatGPT has no idea who you are or what you are talking about. Because of this, it relies heavily on the context provided in the prompts.
Let’s say you’re asking ChatGPT to help draft an email reply to a client facing a technical issue.
Instead of simply asking, “Write a response,” a better approach would be, “Write a polite response to this client using an apologetic tone and outlining the steps to solve the issue.”
Providing context empowers the AI to generate more accurate and helpful responses, as opposed to frustrating generic outputs that don’t apply to your situation.
Be as specific as possible
The more specific your instructions, the closer the output will be to your desired result. This might include specifying the structure, tone, length, or type of output you want.
For example, if you’re asking for an article outline, you could say, “Draft a detailed outline for a 1500-word article on climate change, including an introduction, three main sections with sub-points, and a conclusion, in a conversational tone.”
Additionally, you could tell ChatGPT specific things or opinions you want the article to be based on. “Mention melting ice caps and ocean pollution, and highlight the negative impact of corporation X doing Y.” The more precise your prompts, the more focused and relevant the AI’s response will be.
Tell ChatGPT what you don’t want
ChatGPT is designed to follow instructions. If there are certain topics, words, or types of responses you want to avoid, make sure to state it in your prompt. Otherwise, AI might give you the opposite of what you are looking for.
“Create a captivating marketing campaign that stands out from the competition. Be sure to avoid any clichés or overused buzzwords to maintain a fresh and original tone.”
“Give me 3 ideas for a suspenseful thriller plot without involving any supernatural elements or paranormal occurrences.”
Make use of role prompting
Role prompting involves assigning a specific role to ChatGPT, which can help guide its responses.
For example, instead of asking, “What’s the weather like?” you can say, “Act as an experienced weather forecaster and describe today’s weather.” This not only provides context but also sets the style and tone for the AI’s response.
If you want ChatGPT to write something on your behalf, let it know exactly what it means. “Act as a talented software engineer with 20 years of professional experience, specializing in building secure frameworks for the automotive industry.”
Teach AI your unique style
Here’s a pro tip: you can actually train ChatGPT to mimic your unique writing style. Just give it a piece of your writing as an example, either by copy-pasting or by giving it a link to one of your published articles.
You can say, “Write a blog post about the benefits of meditation in the same style as the provided text.” Or you can just ask it to analyze your article’s language and aim to imitate it in all future writing.
This can help you generate text that’s more in line with your individual style and voice. It will also ensure that the outputs you get will never be exactly the same as someone else gets from a similar prompt.
Don’t expect a perfect result on the first attempt
Iterative refinement is an unavoidable part of effective prompting. You will rarely get the perfect output on your first try, at least not until you master prompt engineering and learn all the tricks by applying them every day.
Instead, consider the AI’s initial response as a draft, and refine your prompt based on that. I often ask ChatGPT to “make it shorter” or even “try to sound more human,” – and I usually get a better result after that.
Patience and continuous refinement can significantly enhance the quality of your results with ChatGPT.
Addressing plagiarism concerns
Plagiarism concerns with AI models like ChatGPT often arise from misunderstandings about how the tool operates. ChatGPT generates responses through predictive analysis based on the patterns and structures it has learned from its vast training data, not by recalling or reproducing specific documents.
However, it’s only logical that if 3 people ask ChatGPT to “write an article about [keyword],” they will get 3 pretty similar articles. That’s why it’s always smart to take steps to ensure the uniqueness of your AI-generated content, based on the concepts and mistakes you learned in this article.
Providing ChatGPT with context, using detailed, personalized prompts, and adding your personal touch will help you create unique outputs nobody else will receive.
That said, it never hurts to run the final piece through a plagiarism checker just to be on the safe side.
Wrap up
To sum up, I think ChatGPT isn’t the problem – it’s an extraordinarily powerful tool that when used correctly, can yield impressive results and make your work easier. Its true potential can be unlocked with well-crafted prompts, and the more you improve your prompts, the better the outputs will be.
Rather than shunning the tool for initial imperfections, embrace it as a learning curve. Experiment, refine your prompts, have some fun with it, and discover how ChatGPT can be an invaluable asset in your professional toolkit.
