The top 1% of any industry is happily leveraging Generative AI to build new businesses, become productive in daily workflows, or simply tinker with an interesting technology out of curiosity. Whatever the reason – one thing is for sure, if you’re not exploring Generative AI tools, there are high chances it may get tough for you to be a part of this top 1%.
The key to making the most of Generative AI tools is to know how to write good prompts. This may change in a few years as AI is evolving to grow smart and may become better at understanding our requirements in the usual conversational language. However, as of now, knowing how to write a prompt is a useful skill to learn to stay updated with this growing leverage of Generative AI tools.
Key takeaways:
- What is a prompt
- Elements of a good prompt
- 5-step process to write a prompt as per best practices
- 3 tips to optimize ChatGPT prompts
What is a Prompt?
Before we dive into the intricacies of prompt engineering, let’s clarify what a ‘prompt’ means in the context of Generative AI.
A prompt is the input or a set of instructions you provide to the AI model to generate a specific output. It acts as a guiding cue, steering the AI towards producing content that aligns with your intent.
The more skilled you are at writing these sets of instructions, the more aligned the output is to what you seek. This is why your prompts should include examples, questions, step-by-step process descriptions, or anything that helps provide the AI model with more context of what you’re seeking from it.
For example, I am facing writer’s block and want to have a unique sentence on ‘color of the sky’ for a poem I am writing. If I write an incomplete sentence, here’s how ChatGPT responds:
As you can see, ChatGPT has given me all the possibilities of what the color of the sky could be. But this is nothing close to poetry, so it is of no use to me. Hence, I will have to improve my prompt to make it specific to poetry.
Let’s try again.
ChatGPT has adopted Sylvia Plath’s melancholic poetry writing and has given an output that could potentially help me come out of my writer’s block.
This is the power of writing good prompts for non-techies – you reduce your cognitive load by hiring a free assistant like ChatGPT so that you can focus on getting results worth executing.
What is a ChatGPT prompt?
ChatGPT prompt is a set of instructions specifically written to get desired output from ChatGPT AI model developed by OpenAI.
It’s important to note that the prompt is the main means users communicate with ChatGPT, guiding the model to generate text based on the input provided.
Elements of a well-written prompt for ChatGPT or Generative AI tools
A well-written prompt has the following elements:
Role: you can define a role that you want the ChatGPT to adopt. For example, it is possible to ask AI models to ‘think like a junior engineer’ and then write the instructions to refine the output that is aligned with this audience.
Instructions: includes a set of tasks that you want the AI model to perform
Context: includes additional information that helps steer the AI model in the direction you want it to follow. This helps make sure your AI model thinks in the way you want it to think so that it produces the desired output. The better you can define the context of your instructions, the more rich your output will be.
Question: you can include a final question to end or summarize your prompt with an ask.
Format: if you want the output to be presented in a certain way, you can specify the format.
Example: including one or more examples of output desired or which follow the format specified helps further refine the output.
5 steps to write an effective prompt for ChatGPT
Writing prompt is a simple, but writing a good prompt is a continuous process of iteration. The good news is that – it’s fun to iterate to get amazing outputs from ChatGPT and know its capabilities.
Please note, that it is better to fact-check ChatGPT’s output on Google before using it because AI models are known to hallucinate. This means they may share content that is not true. Mostly, you can rely on chatGPT for non-fact-based information like asking for recipes, solving code or math problems, getting movie recommendations, etc.
Now, let’s start with simple prompts and we will optimize it with every step by adding various elements of writing a good prompt. You will observe the difference in output as we refine the prompt by adding more contextual elements to it.
Step 1: Write a simple prompt for your use case
Example use case: Drafting a cold email for selling our content writing services.
As a first attempt – write a simple sentence about what you require ChatGPT to do.
Here, I have taken an example of writing a cold email for my content creation services. When I did not mention ‘100 word’ it gave me an extremely long email which I am sure wouldn’t convert at all.
The email shared by ChatGPT is not very specific. Let’s try optimizing the prompt by adding some elements to it.
Step 2: Add a role to your prompt
Let’s see what happens when I update the prompt by assigning a ‘role’ to it. You can use ‘imagine you are a….” and speak about the role you want the AI model to think like. The more you speak about this, the better context you can add to the prompt. The result is that ChatGPT has included my experience and made the cold email better than the one generated in Step 1.
Step 3: Add instructions to the prompt
You can add more instructions to your prompt to make it more specific to your requirements. In our example, we have added an instruction to make the cold email up to ‘150 words’ in Step 2. You can specify further how you want ChatGPT to draft the emails to have better output.
Here’s how I rewrote the prompt to add instructions about writing a good cold email:
Here’s how the output has changed after adding both role and instruction elements in the prompt:
Notice how the output has become better and more casual, yet professionally human [which is how cold emails should be]. But this is still not a usable output for me – because this cold email completely lacks any context about the client I am pitching to.
Step 4: Add context
Now, we will add context to the prompt where I will add more information about the client I am pitching to for my content services. I have also added additional information about brands I previously worked within the ‘role’ section. Here’s how the updated prompt looks like:
Here is the output to this prompt – notice how it has incorporated the webinar aspect and has made the content more specific and personalized to the client:
Step 5: Add a format and/or example
If you already have a format in mind, you can mention the format and also add an example in that format that helps ChatGPT understand and replicate the same for your problem statement.
I have used another prompt to generate a quiz for a YouTube short anime. I specifically wanted this to be in a table format because I wanted to copy and paste the output into an Excel sheet. Here’s how I added the format to make sure ChatGPT produced output as I desired:
Bonus: use reverse prompt engineering to make ChatGPT write your prompts!
The above five-step process is great to get started to explore various capabilities of ChatGPT. But if you’re still confused about writing the prompt and not getting the desired result, here’s a neat trick – how about you ask ChatGPT to create a good prompt for you?
Reverse prompt engineering is simply asking the AI model to write and optimize the prompt for you. You start with a desired outcome or response and work backward to create a prompt that elicits that specific response. It’s a creative approach to leveraging language models like GPT-3 to generate content that fits a predefined goal.
Here’s how I asked ChatGPT to give me a prompt for the same cold email task we were working on in the previous section.
3 tips to optimize your ChatGPT prompt
Optimization is key to getting the best output from ChatGPT.
Remember that ChatGPT responds to the input it receives, so your prompt’s quality significantly influences the generated response’s quality. Adjust and refine your prompts based on the type of information you’re seeking and the context of your inquiry. Once you get a desired response, save the prompts in a document so you can refer to them as required.
Here are three important tips you can explore when you’re not getting the desired output:
1. Explore the chain of thought prompting
Chain of thought prompting is a method that helps you guide the thought process followed by ChatGPT and align it towards what you desire. This technique helps to refine and focus the responses from the model. Here’s how you can use the chain of thought prompting to optimize prompts:
Start with general inquiry:
Write a usual prompt that comes to your mind. Try using elements of the prompt to refine it.
Example Prompt: “Discuss the current state of renewable energy technologies.”
You will observe how ChatGPT will generate a long text with general content.
Ask questions on the generated content:
Just like a student asks a question to a teacher, probe ChatGPT over the generated output in the direction where you want it to produce the content.
Example follow-up prompt: “Now, delve into the advancements in solar energy. What are the key breakthroughs and challenges?”
Write more prompts to do in-depth analysis:
Keep asking questions or ask ChatGPT to expand on generated output that isn’t very clear.
Example follow-up prompt: “Explore the environmental impact of widespread adoption of solar energy. How does it compare to traditional energy sources?”
Write prompts as counter-arguments:
Make ChatGPT think more by sharing counterarguments or asking for alternatives to the generated content.
Example prompt: “What criticisms or limitations are associated with solar energy, and how can these be addressed in future developments?”
Ask ChatGPT to summarize:
When you have reached the desired output, you can ask ChatGPT to summarize everything by considering previous responses.
Example summary prompt: “In summary, provide your perspective on the future trajectory of solar energy technologies and their impact on the energy landscape. Consider previous prompts and their responses shared.”
2. Consider the AI model’s limitation
Understand the limitations of the Generative AI model you’re working with. Tailor your prompts to align with the model’s strengths and weaknesses. For example, ChatGPT is trained till January 2022 data, so it is likely not to give any results which requires knowing current information post the same. Hence, there is no point in writing prompts that involve the use of current affairs data.
3. Monitor Output Consistency
Regularly assess the consistency of AI-generated responses. This helps identify patterns and refine prompts for more reliable outcomes. Here are some tips you can consider:
- Is your model retaining contextual information? – check if your model can incorporate the contextual information you provide on the prompt as well as previous prompts shared. If the model provides a response that seems inconsistent with the established context, you may need to clarify or correct its understanding. If your current prompt builds upon or relates to a previous answer, explicitly refer to that response. This helps reinforce the continuity of the conversation and ensures a coherent flow.
- System Messages for Guidance: use system messages strategically to provide ongoing guidance and maintain the desired behavior. System messages can help establish a consistent tone and context throughout the conversation.
- Are you contradicting your previous prompts? – be cautious about introducing information that contradicts earlier statements in the conversation. If inconsistencies arise, gently guide the model back on track by reinforcing facts or context.
- Use # or ” to separate instruction and context: it is advisable to put the instructions at the start of the prompt and then use ‘#’ of ‘ ” ‘ (double quotes) to separately provide context. This helps ChatGPT not mix up and effectively understand prompts to provide more relevant output.
Frequently Asked Questions (FAQ)
Q1: What is the significance of prompt length?
The length of a prompt can impact the AI’s response. Experiment with different lengths to find the sweet spot for your specific requirements.
Q2: Can I use prompts for creative writing?
Absolutely! Crafting prompts for creative writing is a fantastic application of Generative AI. Be imaginative and specific in your instructions for optimal results.
Q3: How often should I refine my prompts?
Generative AI is an evolving technology, hence you must regularly refine your prompts to keep up with the changes and consistently achieve optimal results. If the AI responses fall short of expectations, iteratively refine your prompts until you achieve the desired outcome.
Q4: Are there specific best practices for prompt engineering?
A4: While there are general guidelines, the effectiveness of prompts can vary based on the specific AI model and use case. Experimentation and iterative refinement are crucial best practices.
Q5: Can I use the same prompt across different Generative AI models?
A5: While some prompts may work across models, it’s advisable to tailor prompts to the specific strengths and nuances of each Generative AI tool for optimal results.
Q6. Does ChatGPT record your prompts?
ChatGPT does not record your prompts and the output unless you instruct it specifically to do so in your prompt. It can go through the previous outputs and you can instruct it to use that if you’re using the chain of thought prompting method. It is also able to remember information within the same chat window. To ensure data security, you must be careful before you share personal information.
Here’s what ChatGPT generates when you ask this question:
Q7. Can I sell ChatGPT content?
As per ChaGPT’s Terms and Content Policy, you possess ownership of the content generated using ChatGPT. This ownership includes the right to reproduce, sell, and merchandise the output, irrespective of whether the content was created under a paid or free plan.
Q8. Why does ChatGPT give incomplete answers?
Sometimes, ChatGPT cannot generate answers because of reaching capacity limits, or maybe you have a bad internet connection. It is also possible your prompt includes some of its limitations – like, maybe your prompt requires access to data post-Jaunuary 2021, which chatGPT is not trained for.
Q9. Does ChatGPT give a new answer every time?
ChatGPT is designed to provide unique answers to the same prompt, although it may still generate similar responses. This diversity in responses is due to the nature of the underlying language model, which is probabilistic and generates outputs based on patterns learned during training. Thus, it will consider your previous prompts, tone, language, or data shared – and that impacts the output, hence increasing the chances of generating a unique output. While the responses may vary, they are all generated to be contextually relevant and coherent.
Q 10. What causes ChatGPT to hallucinate?
ChatGPT is designed to predict the next word to generate text that answers your prompt as per the context and instructions. Thus, sometimes it may pick up a wrong train of thought and generate content that is either contextually or factually incorrect. At present, it is not very great with reasoning. To avoid AI model hallucination, you should include enough context on your prompts and share data from your side to help it work on that instead of coming up with something on its own.
Get started with writing prompts to make the most of Generative AI tools
Generative AI relies on prompt engineering to produce contextually relevant and coherent responses. Effective prompt engineering not only enhances the quality of AI-generated content but also influences the creativity and depth of the responses. Whether you are using ChatGPT or similar tools, mastering the art of crafting prompts is key to unlocking the full potential of these technologies.
- Precision is Key: Craft clear, precise prompts to guide the AI toward your desired output. Reduce any kind of ‘fluff’ or confusing descriptions that may hinder the AI model’s understanding of requirements.
- Experiment and Refine: Regularly experiment with different prompts and iterate based on the AI’s responses.
- Context Matters: Providing context enhances the AI’s understanding and contributes to more accurate and relevant responses.
- Stay Informed: Keep abreast of the evolving field of Generative AI and adjust your prompt engineering strategies accordingly.
Do you have any prompt hacks for ChatGPT or Generative AI tools? We would love to feature your experience on this blog post – email to content@merrative.com
You can subscribe to our newsletter to get notified when we publish new guides – shared once a month!
This blog post is written using resources of Merrative – a publishing talent marketplace that helps you create publications and content libraries.
Get in touch if you would like to create a content library like ours in the niche of Applied AI, Technology, Machine Learning, or Data Science for your brand.
Leave a Reply