Many of us aren’t getting the most out of ChatGPT and it’s not the AI’s fault. Too often, we skip including examples in our prompts, overlook the power of assigning roles to shape its behavior, and leave ChatGPT to guess instead of giving it clear information to work with.
3 Best ChatGPT Prompt Engineering
Why? Because we tend to rely on basic, one-size-fits-all prompts that might work once but fall short of consistent quality. To unlock better results, we need to master the art of creating high-quality prompts otherwise known as prompt engineering. In this guide, we’ll explore 3 essential techniques to elevate your prompt game and get the responses you’re aiming for.
Few-Shot Standard Prompts
Few-shot standard prompts are an upgrade to the basic prompts we’re used to. They include examples of the task at hand. Why bother with examples?
Because they significantly boost your odds of getting the exact output you want by showing the model what success looks like. These prompts have three parts: a task description, a few examples, and the prompt itself essentially the start of a new example that the model completes by filling in the blanks.
Components of Few-Shot Standard Prompts
- Task Description: What you want the model to do.
- Examples: Sample inputs and outputs to guide the model.
- Prompt: The new input you want the model to process.
Let’s say you need to extract airport codes from this text: “I want to fly from Orlando to Boston.” A typical standard prompt might look like this:
Extract the airport codes from this text:Text: “I want to fly from Los Angeles to Miami.”
Airport codes: LAX, MIAText: “I want to fly from Nashville to Kansas City.”
Airport codes: BNA, MCIText: “I want to fly from Orlando to Boston”
Airport codes:
Run this through ChatGPT, and you’ll get “MCO, BOS” formatted just like the examples. Research shows that the specific answers in your examples don’t need to be correct; what matters is the labelspace the range of possible labels (here, airport codes).
Including examples, even with random but plausible labels, helps the model understand the task and format.
Let’s test this with nonsense airport codes:
Extract the airport codes from this text:Text: “I want to fly from Los Angeles to Miami.”
Airport codes: DEN, OAKText: “I want to fly from Nashville to Kansas City.”
Airport codes: DAL, IDAText: “I want to fly from Orlando to Boston”
Airport codes:
Even with random codes, ChatGPT still delivers “MCO, BOS.” The takeaway? Whether your examples are accurate or not, tossing in labels from the labelspace improves results and guides the model on how to structure its response.
Role Prompting
Sometimes ChatGPT’s default mode just doesn’t cut it. That’s where role prompting comes in assigning it a specific persona to tailor its responses. Want to prep for a job interview? Tell ChatGPT to “act as a hiring manager” and layer in details about the role or company. Suddenly, you’ve got a mock interview tailored to any position you’re eyeing.
This trick isn’t limited to interviews. You can transform ChatGPT into a Spanish language tutor, a movie critic, or anything else you need. Start your prompt with “Act as a…” and pile on specifics context is king.
Need ideas? Check out this repository packed with prompts to make ChatGPT a stand-up comedian, doctor, or more. (In another article, I dive deeper into turning it into your personal language buddy.)
Adding Personality and Generating Knowledge
These next two techniques shine when crafting text like emails, blogs, stories, or articles.
Personality in Prompts
By “adding personality,” I mean infusing your prompt with style and descriptors. A style like “in the voice of an expert in [field] with 10+ years of experience” sets the tone, formality, or perspective. Descriptors (think adjectives like “witty” or “sarcastic”) tweak it further.
Take a bland prompt:
“Write a 500-word blog post on how AI will replace humans.” You’ll get something generic. Now spice it up:
“Write a witty 500-word blog post on why AI will not replace humans, in the style of an AI expert with 10+ years of experience. Use funny examples.”
The output? Sharper, livelier, and uniquely flavored. Bonus: that personality makes it tougher for AI detectors to flag (more on fooling detectors in this article).
Generated Knowledge
Want an even richer result? Use the “generated knowledge” approach. First, have ChatGPT whip up useful info on your topic, then feed that into your main prompt. For example:
Step 1:
“Generate 5 facts about why AI will not replace humans.”
Step 2:
# Fact 1
# Fact 2
# Fact 3
# Fact 4
# Fact 5This two-step process grounds the output in specific insights, making it more informed and compelling.
READ ALSO: 15 Winning Pinterest Marketing Strategies To Build Your Brand
Chain of Thought Prompting
Unlike standard prompting, in chain of thought prompting, the model is induced to produce intermediate reasoning steps before giving the final answer to a problem. In other words, the model will explain its reasoning instead of directly giving the answer to a problem.
Why is reasoning important? The explanation of reasoning often leads to more accurate results.
To use chain of thought prompting, we have to provide few-shot examples where the reasoning is explained in the same example. In this way, the reasoning process will also be shown when answering the prompt.
Here’s a comparison between standard and chain of thought prompting.
Source: Google Research
As we can see, the fact that the model was induced to explain its reasoning to solve this math problem led to more accurate results in chain of thought prompting.
Note that chain of thought prompting is effective in improving results on arithmetic, commonsense, and symbolic reasoning tasks.
Tips for Writing Better Prompts
-
Rephrase and Re-ask – If you don’t get the answer you want, tweak your wording and try again. ChatGPT can generate different responses each time.
-
Make Small Adjustments – Slight modifications to your prompt can significantly impact the quality of the response.
-
Maintain Context (When Possible) – As long as you keep the same chat window open, ChatGPT retains awareness of previous messages. However, it may occasionally lose context, so be prepared to restate important details.
-
Fresh Start for New Topics – Opening a new chat resets ChatGPT’s memory, which can be useful when switching to a different topic.
-
Specify Response Length – If you want a concise or detailed answer, mention the desired length. Long responses (500+ words) may sometimes lose coherence.
-
Clarify Misinterpretations – If ChatGPT misunderstands your request, correct it directly or rephrase your prompt for better accuracy.
-
Use Different Wording – If ChatGPT refuses to answer or misinterprets your question, reword it or try using a different approach.
-
Leverage Personas – To get responses from a specific perspective, prompt ChatGPT to “respond as a [role/personality],” such as an expert, teacher, marketer, or journalist.
-
Request Sources or Justifications – If you want citations or factual backing, ask ChatGPT to support its claims with sources.
-
Utilize Custom Instructions – Free users can set permanent instructions in ChatGPT’s settings, ensuring consistent responses without retyping preferences each time.
-
Experiment and Explore – Keep refining your prompts to discover the best ways to interact with ChatGPT.
-
Consider ChatGPT Plus – A subscription provides more processing time and fewer interruptions during longer sessions.
-
Compare with Other AI Tools – Try Gemini (formerly Bard) or Copilot (formerly Bing Chat) to see how different AI models interpret your prompts.
-
Ask for Examples – If ChatGPT’s response is unclear, request multiple examples for better understanding.
-
Echo Back Requests – To ensure ChatGPT understands you, ask it to summarize parts of your input or repeat key details.
-
Know When to Move On – Sometimes, ChatGPT struggles with certain topics. If it’s not working, try another tool or approach.
1. What type of prompts work best with ChatGPT?
The best prompts are clear, specific, and structured. If you need detailed information, provide context and specify the format you want (e.g., “Explain this in simple terms” or “List five key points”). If you need creative content, include style and tone preferences (e.g., “Write a humorous product description for a coffee mug”).
2. How can I adjust the complexity of ChatGPT’s responses?
To simplify a response, ask ChatGPT to “explain this like I’m five” or “summarize this in one sentence.” To make responses more detailed, request an “in-depth explanation,” “technical breakdown,” or “step-by-step guide.” You can also specify the audience level, such as “Write this for beginners” or “Explain this for experts.”
3. How can I adjust the creativity level of ChatGPT’s responses?
For more creative responses, ask ChatGPT to “be imaginative,” “add humor,” or “think outside the box.” You can also request a specific tone, like “Write this as a fantasy story” or “Make this sound like a sci-fi novel.” For factual or structured responses, request “a straightforward and factual explanation” or “a professional, business-style answer.”
4. What do I do if ChatGPT refuses to answer or I don’t like its response?
If ChatGPT declines to answer, try rewording the question or providing more context. If you receive an unsatisfactory answer, ask it to “rephrase,” “expand,” or “focus on a specific aspect.” You can also experiment with different prompts or request multiple versions of the response to compare.
