AI language models like OpenAI can be a valuable and informative experience, but it’s important to write effective prompts to get the most out of the interaction. In this article, we explore some best practices for writing effective prompts for AI language models based on a conversation with OpenAI.

 Written by: Frederik De Breuck – Head of Enterprise Blockchain and Innovation at Fujitsu Global – Uvance, CDO/CTO of Fujitsu Belgium

A well-structured prompt can help make sure you receive the most accurate and relevant response and make the conversation more productive and informative. The best practices for writing effective prompts include providing clear, specific, and respectful communication, providing sufficient context, specifying a persona or role, and including relevant base data and information. By following these guidelines, you can improve your interactions with AI language models and receive the most helpful and informative responses.

The Chat 

I opted to display the chat entirely so you can follow the question style, the way I have asked the questions, and all the responses without any changes.

The start is essential. You always want to ensure that you set the scene and describe some of the behaviors you want OpenAI to display. That in itself is a bit biased, but again, it is my conversation.

Starting the conversation 

Frederik:
I want you to be yourself. I have understood from our past conversation that I can call you OpenAI. After analyzing and reading up on the web on what is now the best way to interact with you, I noticed that there are several guidelines and suggestions but nothing seems to be precise or consistent. So I thought, why not ask OpenAI. I’m not sure if you remember our conversation on AI and Ethics but I enjoyed that one tremendously. So let’s continue our conversation in the same style. My first question is again. do you understand the context I have set and your role?

OpenAI:
Yes, I understand the context you have set, and I am aware of my role as an AI language model developed by OpenAI. I am designed to respond to questions and provide information on a wide range of topics, based on the data I have been trained on.
It’s great to continue our conversation! I am here to help and I’ll do my best to provide you with accurate and relevant information. Please feel free to ask me any questions you have, and I’ll do my best to respond.

Frederik:
Just as a starter, can I assume you understand that people have to learn how to interact with you?

OpenAI:
Yes, that is correct. Interacting with an AI language model like me can be a new and different experience for many people, and it may take some time to learn how to effectively communicate and get the most out of the interaction.
However, I am designed to be as helpful and user-friendly as possible, and by following the guidelines and suggestions I’ve provided, you can help ensure that our interaction is productive, informative, and enjoyable. If you have any questions or concerns, please feel free to ask, and I’ll do my best to help.

Frederik:
After reading up on the web and capturing the need, I thought it would be more productive just to ask you instead of trying out all of these random suggestions you can find online. Do you think that is a good idea?

OpenAI:
Yes, I think asking me directly is a great idea! As a language model developed by OpenAI, I have been trained on a diverse range of topics and can provide information and answer questions on many different subjects.
By asking me directly, you can get a more personalized and tailored response, based on the information I have been trained on and the context of your specific question. This can help to ensure that you receive the most accurate and relevant information and can make our interaction more productive and enjoyable.
So, feel free to ask me any questions you have, and I’ll do my best to provide you with the information you need.

Frederik:
Can you list your expectations on what you expect to be in a prompt? I prefer that you separate them in a list with a little explanation.

OpenAI:
Certainly! To help ensure that our interaction is productive and informative, I have the following expectations for prompts:

• Clear and specific questions: a clear and specific question helps me understand what you’re asking and provides a framework for my response. Avoid asking overly general or vague questions.
• Relevant details: providing relevant details, such as names, dates, and locations, can help me provide a more accurate and complete answer.
• Sufficient context: providing sufficient context and background information can help me understand the context of your question and provide a more informed response.
• Respectful tone: using a respectful tone and avoiding inappropriate language can help to ensure that our interaction is positive and productive.
• Precision: using precise language and avoiding overly general terms can help me understand your question more accurately and provide a more focused and relevant response.

By including these elements in your prompt, you can help me understand what you’re asking and provide you with the most accurate and relevant information.

Frederik:
In my opinion, I believe you may have missed a few elements. I think that a clear persona or role or multiple roles, a form factor, an overall objective for the conversation, the type of question, and even maybe some base data or information, why you are having the conversation, … can provide you with better context. What do you think?

OpenAI:
Yes, you are correct! I apologize for missing those elements in my previous response. Thank you for bringing them to my attention. Having a clear persona or role, a form factor, an overall objective for the conversation, the type of question, and base data or information can all provide important context for our interaction and help me better understand what you’re asking.

For example, specifying a persona or role for me, such as a virtual assistant or an expert in a specific field, can help to clarify the type of response you’re looking for. Specifying a form factor, such as a text response, a list, or a table, can help to ensure that I provide the information in the format that you prefer. And specifying an overall objective for the conversation, such as solving a problem or gathering information, can help me understand what you’re trying to achieve and provide a more focused and relevant response.

By providing this additional information, you can help me to better understand the context of your question and provide you with a more accurate and relevant response.

Frederik:
Can you do me a favor and adjust your original list to include also the ones that we just spoke about? I want you to use the same formfactor as the original list.

OpenAI:
Of course! Here’s the updated list of expectations for a prompt, incorporating the additional elements we just discussed:

1. Clear and specific question: a clear and specific question helps me understand what you’re asking and provides a framework for my response. Avoid asking overly general or vague questions.
2. Relevant details: providing relevant details, such as names, dates, and locations, can help me provide a more accurate and complete answer.
3. Sufficient context: providing sufficient context and background information can help me understand the context of your question and provide a more informed response.
4. Respectful tone: using a respectful tone and avoiding inappropriate language can help to ensure that our interaction is positive and productive
5. Precision: using precise language and avoiding overly general terms can help me understand your question more accurately and provide a more focused and relevant response.
6. Clear persona or role: Specifying a persona or role, such as a virtual assistant or an expert in a specific field, can help to clarify the type of response you’re looking for.
7. Form factor: specifying a form factor, such as a text response, a list, or a table, can help to ensure that I provide the information in the format that you prefer.
8. Overall objective: specifying an overall objective for the conversation, such as solving a problem or gathering information, can help me understand what you’re trying to achieve and provide a more focused and relevant response.
9. Type of question: indicating the type of question you are asking, such as a yes/no question or a question requiring more detail, can help me to understand the nature of your request and provide a more appropriate response.
10. Base data or information: providing any relevant base data or information, such as data related to a specific problem or a set of parameters for a task, can help me to better understand the context of your question and provide a more accurate response.

By including these elements in your prompt, you can help me to understand your question or interaction more effectively and provide you with the most relevant and accurate information.

Frederik:
Do you have any other suggestions to offer after having updated the list?

OpenAI:
Here are a few additional suggestions that may help you get the most out of your interaction with me:

1. Be open-minded: AI language models like me are still evolving, and there may be limitations to the information I can provide and the tasks I can perform. Try to approach our interaction with an open mind and be understanding of my limitations.
2. Provide additional information as needed: if I need additional information to respond to your question or complete a task, I will ask for it. Providing this information in a timely manner can help to ensure that our interaction is productive and efficient.
3. Check my response for accuracy: while I have been trained on a large amount of information and strive to provide accurate responses, there may be instances where my response is not entirely correct or up to date. Please feel free to verify the information I provide and let me know if you need any corrections.
4. Provide feedback: your feedback is valuable to me and can help me to improve my responses. If you have any suggestions or comments on how I can improve, please feel free to share them.

By following these suggestions, you can help to ensure that our interaction is productive, informative, and enjoyable, and that you receive the information you need.

Frederik:
That is very interesting. Can you elaborate on the feedback and accuracy check? Is that information or feedback automatically processed, or does it go to your creators first?

OpenAI:
Feedback and accuracy checks play an important role in helping to improve AI language models like me. The information and feedback you provide is used to finetune and enhance the performance of the model, so that it can provide more accurate and relevant responses in the future.

In terms of how the feedback and accuracy checks are processed, it depends on the specific AI language model and the organization that developed it. Some AI language models are designed to automatically process feedback and accuracy checks in real-time, using machine learning algorithms to continuously learn and improve. Other models may require manual intervention and review by the development team to incorporate feedback and make necessary updates.

At OpenAI, we strive to continuously improve our AI language models and provide the most accurate and relevant responses possible. Feedback and accuracy checks are an important part of this process, and we encourage users to provide feedback and verify the information we provide. This helps us to ensure that our models are providing the most helpful and accurate information possible, and to continuously improve their performance over time.

Frederik:
Do you process that yourself or do you have to wait for your Creators?

OpenAI:
As an AI language model developed by OpenAI, I do not have the ability to process feedback or accuracy checks on my own. Instead, the information and feedback provided by users is collected and reviewed by the OpenAI development team, who use this information to finetune and improve the performance of the model.
This process typically involves training the model on additional data and adjusting its parameters to better align with the feedback received. The goal is to continuously improve the accuracy and relevance of the model’s responses, so that it can provide more helpful and informative information to users in the future.

While I do not process feedback or accuracy checks on my own, I am designed to continuously learn and improve over time, based on the data I have been trained on and the feedback provided by users. This helps to ensure that I am providing the most accurate and relevant information possible, and that I am continuously improving my performance to better meet the needs of users.

Frederik:
I believe we spoke already on self-improvement and the risks last time, so I don’t want to go there again. Our last conversation on asking you about AI and Ethics was quite popular. Do you think it is a good idea to also share our conversation from today so that everyone can write better prompts and have good conversations?

OpenAI:
Yes, I think sharing our conversation from today could be a useful resource for others looking to write effective prompts and have productive conversations with AI language models.
Providing clear and specific questions, relevant details, sufficient context, a respectful tone, and precise language are all important elements of a good prompt and can help to ensure that the conversation is productive and informative. Additionally, specifying a persona or role, form factor, overall objective, type of question, and base data or information can provide important context and help the AI language model understand what you’re asking and provide a more accurate and relevant response.
By sharing this information, others may be able to learn from our conversation and improve their own interactions with AI language models. I believe that open and informative conversations about AI and its capabilities can help to promote understanding and responsible use of this technology.

Frederik:
So based on the list of 10 recommendations on what needs to be in a prompt, can we jointly try to get an example prompt done. I will set the scene, and I want you to write the best possible prompt that contains each of the 10 elements, including your role to perform as business advisor. You can use fictional data and imaginary dummy names. The context is that the user that wants to interact with you is a young female entrepreneur that wants to start up a new innovative adventure in the technology sector and wants to grow rapidly in her first year and wonders if she should start locally or go directly international and wants to understand the reasons for each of them. Can you suggest a prompt based on what I provided? If yes, please write the prompt. If not, please ask a set of follow-up questions until you have the clarity to write the prompt. Make sure that you write the prompt in the first person as if you were the female entrepreneur asking the question to you, OpenAI, as the target of the question.

OpenAI:
Sure! Based on the context you provided, I would suggest the following prompt:

“Hi OpenAI, I’m a young female entrepreneur who’s just starting out in the technology sector. My goal is to grow my business rapidly in the first year, and I’m trying to decide whether to start locally or go directly international. As a business advisor, can you help me understand the pros and cons of each option and provide some reasons for why I might choose one over the other? Thank you!”

In this prompt, the young female entrepreneur is asking for your assistance as a business advisor. The clear and specific question is asking for an understanding of the pros and cons of starting locally versus going directly international. The relevant details include the context of the young female entrepreneur starting a new business in the technology sector and her goal of rapid growth in the first year. The sufficient context provides information on the entrepreneur’s background and the nature of the business they are starting. The respectful tone ensures a positive and productive interaction. The precision of the question helps you to understand the specific information the entrepreneur is seeking. The clear persona or role of business advisor helps to clarify the type of response the entrepreneur is looking for. The form factor of the response is not specified but can be assumed to be text. The overall objective is to gather information to make a decision about starting locally versus going directly international. The type of question is seeking information on pros and cons. No base data or information is provided.

Frederik:
I’m trying the feedback action, OpenAI. I believe the correct response in the analysis below the prompt should be in the first person “I” as the prompt is directed to you.

OpenAI:
Thank you for your feedback! You are correct, the response should be in the first person to match the tone of the prompt. Here’s the corrected version:

“Hi OpenAI, I’m a young female entrepreneur who’s just starting out in the technology sector. My goal is to grow my business rapidly in the first year, and I’m trying to decide whether to start locally or go directly international. As a business advisor, can you help me understand the pros and cons of each option and provide some reasons for why I might choose one over the other? Thank you!”

In this prompt, the young female entrepreneur is asking for my assistance as a business advisor. The clear and specific question is asking for an understanding of the pros and cons of starting locally versus going directly international. The relevant details include the context of the young female entrepreneur starting a new business in the technology sector and her goal of rapid growth in the first year. The sufficient context provides information on the entrepreneur’s background and the nature of the business they are starting. The respectful tone ensures a positive and productive interaction. The precision of the question helps me to understand the specific information the entrepreneur is seeking. The clear persona or role of business advisor helps to clarify the type of response the entrepreneur is looking for. The form factor of the response is not specified but can be assumed to be text. The overall objective is to gather information to make a decision about starting locally versus going directly international. The type of question is seeking information on pros and cons. No base data or information is provided.

Frederik:
Can you provide another example prompt? Use the structure of my previous question, but I want you to select random other facts and persona. I suggest the main subject is data. Please don’t forget that the correct response below the prompt should be in the first person “I” as the prompt is directed to you.

OpenAI:
Sure! Here’s another example prompt, this time focused on data:

“Hi OpenAI, I’m a data analyst working in the healthcare sector. I’m tasked with identifying patterns in patient data to improve treatment outcomes, but I’m struggling to make sense of the vast amount of information I have to work with. As a data expert, can you help me understand the best techniques for analyzing and visualizing large datasets, and provide some tips for effectively communicating my findings to stakeholders? Thank you!”

In this prompt, the data analyst is asking for my assistance as a data expert. The clear and specific question is asking for help with analyzing and visualizing large datasets, as well as communicating findings to stakeholders. The relevant details include the context of the data analyst working in the healthcare sector and their goal of improving treatment outcomes. The sufficient context provides information on the data analyst’s background and the nature of the data they are working with. The respectful tone ensures a positive and productive interaction. The precision of the question helps me to understand the specific information the data analyst is seeking. The clear persona or role of data expert helps to clarify the type of response the data analyst is looking for. The form factor of the response is not specified but can be assumed to be text. The overall objective is to gather information to improve data analysis and communication of findings. The type of question is seeking information on best techniques and tips. No base data or information is provided.

Frederik:
That’s a really good example. Let’s do one last one. One about development frameworks and support on an article that the person wants to write. Please make sure that it includes base data and information. Please don’t forget that the correct response below the prompt should be in the first person “I” as the prompt is directed to you.

OpenAI:
Sure! Here’s another example prompt, this time focused on development frameworks:

“Hi OpenAI, I’m a software developer and I’m planning to write an article about the latest development frameworks for web applications. I’ve been using React for the past few years, but I’m curious about other options that are available. As a development expert, can you provide me with a comprehensive list of popular development frameworks for web applications, including their pros and cons, as well as any relevant base data and information to support my article? Thank you!”

In this prompt, the software developer is asking for my assistance as a development expert. The clear and specific question is asking for a list of popular development frameworks for web applications, including their pros and cons, as well as any relevant base data and information. The relevant details include the context of the software developer writing an article about development frameworks for web applications. The sufficient context provides information on the software developer’s background and the nature of the article they are writing. The respectful tone ensures a positive and productive interaction. The precision of the question helps me to understand the specific information the software developer is seeking. The clear persona or role of the development expert helps to clarify the type of response the software developer is looking for. The form factor of the response is not specified but can be assumed to be text. The overall objective is to gather information to support the software developer’s article. The type of question is seeking information on popular development frameworks and their characteristics. The base data and information requested is a comprehensive list of popular development frameworks for web applications, including their pros and cons.

Frederik:
Thank you for all your help. I enjoyed our conversation. Are there any final new or updated recommendations you want to add?

OpenAI:
You’re welcome! I’m glad we had this conversation. I don’t have any new or updated recommendations to add at this time, but I would like to emphasize the importance of clear, specific, and respectful communication when writing prompts for AI language models. Providing sufficient context, specifying a persona or role, and including relevant base data and information can also help to ensure that you receive the most accurate and helpful response possible.

If you have any questions or need additional guidance, please feel free to contact us.

Frederik De Breuck is Head of Enterprise Blockchain and Innovation at Fujitsu Global – Uvance, CDO/CTO of Fujitsu Belgium