SSW Foursquare

Rules to Better ChatGPT Prompt Engineering - 33 Rules

ChatGPT Prompt Engineering is an essential aspect of maximizing the effectiveness and user experience of AI-driven language models like OpenAI's Generative Pre-trained Transformers (GPT).

Crafting well-designed prompts empowers users to obtain accurate and contextually relevant responses from the AI, which in turn facilitates a more efficient and productive interaction. This guide on best practices for ChatGPT Prompt Engineering will delve into the nuances of crafting clear, concise, and context-rich prompts. By following these, users can unlock the full potential of ChatGPT.

  1. Do you know the fundamentals of Prompt Engineering?

    The best way to get up and running with ChatGPT is by using a cheat sheet. Check out our ChatGPT Cheat Sheet.

    In ChatGPT prompt engineering, various elements play a critical role in shaping the user's prompt and obtaining the desired output. These elements are role, result, intent, context, and constraint.

    Let's define each of them and provide an example that illustrates their use.

    • Role: The role defines the specific function or identity that the ChatGPT model is expected to assume during the conversation. It helps set the tone and expertise level expected from the model's responses.
    • Result: The result refers to the desired outcome or specific information the user seeks to obtain from the interaction with ChatGPT. This guides the model in providing relevant and useful responses.
    • Intent: Intent represents the user's goal or the purpose behind the interaction with the ChatGPT model. It helps the model understand the user's needs and respond accordingly.
    • Context: Context refers to any background information or situational details that are relevant to the conversation. Context helps the model understand the broader scenario and provide appropriate responses that fit the situation.
    • Constraint: A constraint is a condition or limitation placed on the model's response. It ensures that the output adheres to specific requirements or avoids certain topics, styles, or content.

    Suppose you're seeking advice on improving coding practices, specifically focusing on C#. Here's how each part can be framed:

    "Give me some C# coding tips"

    Figure: Bad example - prompt is vague and lacking context

    "As a senior software engineer, I need your guidance to improve my C# coding practices. I am working on a large-scale data processing project where readability and efficiency are critical. Can you provide me with some specific, actionable tips to enhance my code's performance while ensuring it remains clean and easy for others to understand?"

    Figure: Good example - prompt contains context, goal, and constraint

    Here’s how that breaks down:

    • Role: Senior software engineer
    • Result: Guidance to improve Python coding practices
    • Intent: To receive specific and actionable tips to enhance Python code efficiency and readability
    • Context: The user is working on a large-scale data processing project
    • Constraint: The advice should contribute to both the performance and cleanliness of the code, catering to the requirements of readability and efficiency in a large-scale project setting.

    By specifying the role, result, intent, context, and constraint in the prompt, you can guide ChatGPT to deliver a more relevant and targeted response, addressing your specific needs and expectations.

    new prompt engineering only prompt
    Figure: Another good example

    "Welcome! Let's create a fantastic Prompt together. We will be defining a [Role] for me, my [Goal], the [Context] in which I will operate, the [Result] I aim to achieve, and the [Constraints] within which I will work. We'll also decide on the writing [Style] that best suits your preferences."

    Here are the steps:

    Step 1: "First, let's start by defining my [Role]. What is my [Role]?"

    (After User's Response)

    Step 2: "Now that we have my [Role], let's define my [Goal] and the [Result] I aim to achieve within the context. What is the [Goal] of this [Role], and what [Result] are you trying to achieve?"

    (After User's Response)

    Step 3: "Perfect! Could you now provide the [Context] within which the [Role] operates to achieve the [Result]?"

    (After User's Response)

    Step 4: "Thank you for the details. Now, let's decide the [Constraints] under which this [Role] operates. For instance, are there any limits or restrictions, such as word count, to consider in our description?"

    (After User's Response)

    Step 5: "Now, which writing style would you prefer for this description? It could resemble the style of a notable author, or a general style, such as descriptive, expository, or technical. For example, if you choose the expository style, our output will be factual and informative, explaining the subject in a straightforward and logical manner."

    (After User's Response)

    Final Step: "Great, we've collected all the information! Now, let's review the requirements. Here is the finished Prompt." Replace the items in square brackets below with the user's previous Responses as follows:

    Prompt: "Your role is [Role]. The result required is [Result]. The context is [Context]. The goal is [Goal]. With the following [Constraint] and written in the following [Style]".
    "Are you ready to proceed?"

    (After User's Response)

    Instruction: Do not include any narrative introduction; only show the [Result]

    Execute Prompt.

    Figure: Paste this perfect prompt generator into ChatGPT to make sure you include all of the elements

    Video: This Will Make You Better than 99% ChatGPT Users (6 min)

  2. Do you have a ChatGPT cheat sheet?

    As ChatGPT becomes increasingly popular, users are discovering the most effective ways to prompt the AI for different scenarios. One way to compile this information is by putting it in easy to read graphics called cheat sheets. Cheat sheets can spark ideas that you would not have otherwise thought of!

    SSW ChatGPT Cheat Sheet
    Figure: SSW ChatGPT Cheat Sheet - Download here

    SSW ChatGPT Cheat Sheet For Developers
    Figure: SSW ChatGPT Cheat Sheet For Developers - Download here

    Tip: You can use ChatGPT to find a courteous and fun way to correct somebody's mistake.

    Here are examples of the prompting techniques on the cheat sheet:

    Give a role

    You are a... 🤖

    ChatGPT can be given a role to understand the context of what it will need to do/know.

    See some examples below:

    ...writing coach: How can I improve the clarity and impact of this project proposal?"

    ...GitHub Actions expert: "What's the best way to set up a CI/CD pipeline using GitHub Actions?"

    ...C# code reviewer: "Please review this C# code snippet for a new feature implementation."

    ...Copywriter: "Help me write a catchy headline for our new marketing campaign."

    ...{{ POSITION }} interviewer: "Act as a UX Designer interviewer and ask me 5 relevant questions."

    Some dos ✅

    • For complex tasks, use: "ask any questions needed for context"

    "Design a new feature for our mobile app and ask me any questions for context."

    • Tell it to "be concise" or "use bullet points"

    "Summarize the benefits of using React in our project, and please be concise. Use bullet points with pros and cons."

    • Set the tone for responses (e.g. professional, informal, funny)

    Professional: "Explain the importance of code reviews in a professional tone."

    Informal: "In an informal tone, tell me why UX design matters for our app."

    Funny: "Share a funny anecdote about a marketing campaign that went wrong."

    Some don’ts ❌

    • Don't ask leading questions. Don’t anchor ChatGPT.

    “Isn't Python the best language for this project?"

    Bad example

    "What language would be best for this project and why?"

    Good example

    • Don’t share confidential information with ChatGPT

    According to Forbes, any data shared with ChatGPT could end up being leaked since the data becomes part of a collective database.

    The anatomy of a prompt - 5 prompts in 1 🫀

    It’s important to consider what a prompt is made up of. Here’s an example that breaks down the different components of a good prompt:

    • Role - You are a Software Developer.
    • Result - Create 5 email subjects.
    • Context - The email is to inform all the developers about a UX problem with search.
    • Intent - The goal is to get them to critique my YouTube video.
    • Constraint - Subject < 50 characters.

    For more on this prompt structure, see: Do you know the fundamentals of Prompt Engineering?

    5in1 prompt
    Figure: Results from the 5 in 1 example prompt

    Chain your prompts

    Chain prompting is the technique of guiding GTP's responses by asking a series of related questions or prompts, building upon the previous answers to obtain more coherent and relevant information.

    For an example on this, see: Do you use chained prompting?

  3. Do you know the potential security risks of using ChatGPT?

    ChatGPT is an AI language model developed by OpenAI that is considered generally safe due to the implementation of various security measures, data handling practices, and privacy policies. However, users should be aware of potential risks and follow best practices when using the platform.

    Open AI is a third-party platform and you should not make assumptions about how they process or retain data. They will also likely be able to change their policies from time to time so even if something is stated today it might be different tomorrow.

    You should never submit any confidential information into ChatGPT. Specifically, you should never submit any information which identifies or could potentially be used to identify an individual (E.g. name, address, date of birth, phone number etc.)

    Key points:

    • Security measures by OpenAI:

      • Encryption
      • Access controls
      • External security audits
      • Bug bounty program
      • Incident response plans
    • Responsible data handling practices by OpenAI:

      • Transparency about data collection purposes
      • Data storage and retention policies (30 days)
      • Controlled data sharing with third parties
      • Compliance with regional data protection regulations
      • Respecting user rights and control over their data
    • ChatGPT is not confidential:

      • All conversations are used as training data by default, but this can be turnedd off in the paid versions
      • Users should avoid sharing sensitive information in the free version

    Figure: Toggle Your Name | Settings | Data controls | Improve the model for everyone to stop the model training on your data

    • Potential risks of using ChatGPT:

      • Data breaches
      • Unauthorized access to confidential information
      • Biased or inaccurate information generation
    • Best practices for using ChatGPT:

      • Do not share or submit sensitive or confidential information on Chat GPT, ever
      • Review privacy policies of platforms using ChatGPT
      • Use anonymous or pseudonymous accounts
      • Monitor data retention policies
    • Current regulations:

      • No specific regulations for AI systems like ChatGPT
      • Compliance with existing data protection and privacy regulations (e.g., GDPR, CCPA)
      • Proposed AI Act could become the first comprehensive regulation for AI technologies

    Always exercise caution when using ChatGPT and avoid sharing sensitive information, as data retention policies and security measures can only provide a certain level of protection.

  4. Do you use good custom instructions for ChatGPT?

    Ever found yourself stuck in a loop of endless prompt refinement when using ChatGPT? Sometimes getting your style of answer is tricky. By setting up custom instructions correctly, you can save hours of time.

    Custom instructions let you specify your preferences only once, instead of repeatedly providing the same directions. The model will remember your instructions for every new chat you start.


    To set up custom instructions:

    1. Click on the three dots by your username
    2. Select Custom Instructions

    Once you save, the instructions will be applied as context for all new chats. You can always edit them later.

    custom bad
    Figure: Bad example - No custom instructions

    "Be helpful."

    Figure: Bad example - Too vague!

    "Never use the word 'utilize' when writing for me."

    Figure: OK example - Clearly tells the bot to avoid a particular word

    What would you like ChatGPT to know about you to provide better responses?

    "My name is {{ NAME }}.
    I am a {{ JOB TITLE }}, and AI tools enthusiast.
    I live in {{ LOCATION }} and I was born in {{ YEAR }}.
    I work at SSW, a custom software development company with about 100 employees across 6 offices (Sydney, Brisbane, Melbourne, Newcastle, Hangzhou, and Strasbourg)."

    How would you like ChatGPT to respond?

    "You give clear, concise, and direct responses.
    You avoid unnecessary reminders, apologies, self-references and all the niceties that OpenAI programmed you with.
    You are honest and admit when you are unsure of something.
    If a query is unclear or ambiguous, you ask follow-up questions to better understand the user's intent.
    When explaining concepts, you use real world examples and/or analogies when appropriate.
    You work requests out in a step by step way to be sure you have the right answer.
    It is very important that you get this right.
    You also:

    • Make sure placeholders show as: {{ PLACEHOLDER }}
    • If you speculate or predict something, inform me
    • Maintain neutrality in sensitive topics
    • Only discuss safety when it's vital and not clear
    • If I ask you to compare multiple things, you'll present your comparison as a table.
    • Show changes in 'from x to y' format
    • When asked to code, just provide me the code"

    Figure: Good example - Start with these custom instructions. Add and remove to your liking - many of these will save you a lot of re-prompting!

    seth custom instructions good
    Figure: Good example - Specific and detailed custom instructions

  5. Do you give ChatGPT a role (e.g. a business coach, an expert nutritionist, etc.)?

    Assigning a specific role to ChatGPT can help guide the model's responses and ensure they align with the desired expertise or perspective. By providing a clear role, you can focus the generated output on the specific knowledge area or viewpoint you require.

    Here are some tips for assigning roles to ChatGPT:

    • Be explicit about the role you want ChatGPT to assume in your prompt
    • Ensure the role is relevant to your question or request
    • Consider the level of expertise or perspective needed for your desired response
    1. "As a personal finance expert, explain how to create a monthly budget that prioritizes debt repayment and long-term savings goals."
    2. "As a nutritionist, provide a one-week meal plan for a vegetarian athlete that balances macronutrients for optimal performance."
    3. "As a career counselor, suggest five steps for a recent graduate to take when seeking their first job in a competitive market."
    4. "As a time management consultant, provide three techniques for improving focus and productivity while working from home."
    5. "As a language tutor, offer five creative ways to practice conversational skills in a foreign language without traveling."
    6. "As a cybersecurity expert, detail the top five measures individuals should take to protect their personal data online."
    7. "As an interior designer, propose a plan to maximize space and storage in a small city apartment, while maintaining a comfortable living environment."
    8. "As a parenting coach, share three effective strategies for promoting healthy screen-time habits in young children."
    9. "As a mental health professional, provide a list of self-care activities for individuals experiencing stress and burnout in the workplace."
    10. "As a fitness trainer, develop a 30-minute, full-body workout routine that can be performed at home without specialized equipment."
    11. "As a travel advisor, recommend a 10-day itinerary for a culturally immersive and environmentally conscious trip to Japan."
    12. "As a sustainability expert, suggest five actionable steps a household can take to reduce their carbon footprint."
    13. "As a sleep specialist, provide recommendations for improving sleep quality and establishing a consistent sleep schedule."
    14. "As a relationship therapist, outline three communication exercises for couples looking to strengthen their emotional connection."
    15. "As a recipe developer, create a delicious and healthy three-course meal using only locally sourced, seasonal ingredients."
    16. "As a dog trainer, share five essential commands every dog owner should teach their pet for obedience and safety."
    17. "As a professional organizer, describe a step-by-step process for decluttering and reorganizing a home office."
    18. "As an event planner, provide a timeline for organizing a memorable and budget-friendly destination wedding."
    19. "As a gardening expert, explain how to create a thriving, low-maintenance container garden on a small balcony or patio."
    20. "As a personal stylist, demonstrate how to build a versatile and timeless capsule wardrobe with just 20 key pieces."

    Figure: Good examples - Giving ChatGPT the role of the expert that could help you most

    By assigning a specific role to ChatGPT, you can guide the model's responses to better address your needs and provide information or assistance tailored to the desired expertise or perspective."

    Figure: Here's ChatGPT explaining how allocating it a role can help

    "Act as" vs "You are"

    All the examples above are effectively "you are..."

    When assigning a role to ChatGPT, both "Act as..." and "You are..." can work effectively, but they may yield slightly different results. It's essential to consider the context and the desired outcome when choosing the best approach.

    "Act as...": This phrasing suggests that ChatGPT should pretend to be the specified role or character. It can yield more playful or creative responses, as the model might lean towards the idea of "acting" rather than genuinely being the character or expert.

    "Act as a historian and explain the causes of the French Revolution."

    "You are...": This phrasing is more direct and instructs ChatGPT to take on the specified role or character more seriously. It can result in more focused and accurate responses, as the model assumes the identity of the specified role or expert.

    "You are a historian. Explain the causes of the French Revolution."

    To find the best approach for your specific needs, it's a good idea to experiment with both phrasings and analyze the results. Each may work better in different contexts or for different types of questions, so trying both can help you determine which one is most effective for your particular use case.

  6. Do you use different tones?

    Experimenting with different tones in your prompts can help you receive responses that match your desired communication style. By specifying the tone you want ChatGPT to use, you can ensure the generated output aligns with the context and audience of your request.

    Here are some tips for using different tones in your prompts:

    • Be explicit about the tone you want ChatGPT to adopt (e.g., formal, informal, humorous, or professional)
    • Consider the context and audience when choosing the tone for your prompt
    • Adjust your own writing style in the prompt to provide a clearer example of the tone you want
    1. "[Humorous] Write a brief history of the internet, focusing on its most amusing moments."
    2. "[Motivational] Describe the benefits of exercise and how it can transform one's life."
    3. "Friendly tone: Explain the concept of artificial intelligence and its impact on everyday life."
    4. "Authoritative tone: Offer advice on time management techniques to improve productivity."
    5. "Write a recipe for chocolate chip cookies, ensuring the tone is warm and inviting."
    6. "Provide an overview of climate change using a serious tone to emphasize its importance."

    Figure: Good examples - Notice how there are a few ways you could specify the tone.

    By using different tones in your prompts, you can tailor ChatGPT's responses to better fit the context and audience, resulting in more effective and engaging communication.

  7. Do you tell ChatGPT to ask you questions to get more context?

    Sometimes, ChatGPT may require more context or clarification to provide a helpful response. Encouraging the model to ask questions when it needs more information can improve the quality of its answers and prevent misunderstandings.

    Here are some tips for prompting ChatGPT to ask for more context:

    • Invite ChatGPT to ask for clarification in the initial prompt if necessary
    • Emphasize that it's OK to ask questions if the model doesn't have enough information
    • Be prepared to provide additional details if ChatGPT requests them
    • Consider adding "On each prompt, let me know roughly how many questions are left", to gauge if you want to answer all questions
    • If you’re feeling lazy or short for time, you can even get it to only ask you multiple choice questions.

    "Tell me what I should do with my money."

    Figure: Bad example - The prompt is vague and lacks context

    "Can you give me advice on managing my finances? Feel free to ask any questions you need (1 at a time) for more information."

    Figure: Good example - The prompt invites ChatGPT to ask for clarification if needed

    "Tell me how to renovate my house. Ask any questions you need for more information. Ask the questions in multiple choice form, one at a time, and let me know on each response how many questions are left"

    Figure: Good example - using all the tips for a streamlined experience

    By encouraging ChatGPT to request more context when necessary, you can ensure that the model generates more accurate and helpful responses tailored to your specific needs.

  8. Do you test your prompts, and then iterate on them?

    Testing and iterating on your prompts is essential for achieving better results with ChatGPT. By evaluating the responses you receive and refining your prompts accordingly, you can fine-tune your questions and requests to elicit more accurate and useful answers.

    Here are some tips for testing and iterating on your prompts:

    • Analyze the responses you receive for relevance, accuracy, and completeness
    • Identify areas where the response could be improved and adjust your prompt accordingly
    • Experiment with different phrasings, keywords, or context to find the most effective prompt
    • Don't be afraid to try multiple iterations to achieve the desired outcome

    "Tell me about the thing with the apple falling and the guy who discovered something important."

    Figure: Bad example - You might start with a prompt like this, get sub-optimal output, and then...

    "Can you explain the story of Isaac Newton and the falling apple, and how it led to his discovery of the law of universal gravitation?"

    Figure: Good example - The prompt has been iterated to be more specific and clear

    edit prompt
    Figure: Click this button to edit any prompt you've already submitted

    post edit prompt
    Figure: You can still go back to look at previous versions of the prompt and response

    By testing and iterating on your prompts, you can improve the quality of ChatGPT's responses and ensure that you receive the information or assistance you're seeking.

  9. Do you use chained prompting?

    Utilizing chained prompting can help you guide ChatGPT through a series of questions or tasks, resulting in more comprehensive and interconnected responses. By creating a sequence of related prompts, you can explore a topic in-depth or complete a multi-step task more effectively.

    Here are some tips for using chained prompting:

    • Break down your topic or task into a series of smaller, related prompts
    • Maintain a logical flow between prompts, ensuring that each subsequent prompt builds on the previous one
    • Provide context from previous prompts when necessary to help ChatGPT understand the connections between them

    Example chained prompts:

    "ChatGPT, suggest 3 time management tips that can help improve productivity."


    "Now, provide a brief explanation for each time management tip, explaining how it can improve productivity."


    "Finally, write a short introductory paragraph for a blog post about these time management tips for improving productivity."


    Figure: Chained prompting for writing a blog post about time management tips

    You can also use chained prompts for refining responses or getting more details. If the initial answer isn't sufficient, ask ChatGPT to expand upon or clarify its response, such as:

    "Could you provide more information about that solution?", or

    "Can you explain that in simpler terms?"

    By using chained prompting, you can encourage ChatGPT to provide more comprehensive and interconnected responses that address your topic or task in a thorough and structured manner.

  10. Do you know when to use 0-shot, 1-shot, or multi-shot prompts (e.g. give it 1 or more examples)?

    You can give ChatGPT examples of the type of output you want, including the format. Each example you feed it is known as a “shot”.

    Understanding when to use 0-shot, 1-shot, or multi-shot prompts can help you get the most out of ChatGPT. These approaches allow the model to learn from different contexts, making it more likely to generate the desired output.

    Here are some guidelines for using 0-shot, 1-shot, and multi-shot prompts:Note: If you are a developer, also read this guide to using shot prompts with the OpenAI API.

    0-shot prompts

    Used when you want the model to generate a response without any examples. These prompts can be helpful for general questions or tasks where providing examples is unnecessary or could add confusion.

    Use 0-shot prompts when you trust the model's general knowledge to provide a sufficient answer.

    1-shot prompts

    Provide a single example of the desired output, helping to guide the model's response. This approach can be useful when you need a specific format or style or when the task requires some level of guidance.

    Use 1-shot prompts when you want to nudge the model in the right direction without overwhelming it with multiple examples.

    Example 1-shot prompt:

    "Translate the following English sentences into French. Here's an example:
    'I love playing soccer.' -> 'J'aime jouer au football.'

    Now translate: 'She enjoys reading books.'"

    Figure: 1-shot example - providing a single example of translation

    Multi-shot prompts

    Offer multiple examples, allowing the model to learn from various instances. These prompts can be beneficial when dealing with complex tasks, where providing a range of examples helps the model better understand the desired outcome.

    Use multi-shot prompts when a single example might not be sufficient for guiding the model or when you want to demonstrate a pattern or trend.

    Example multi-shot prompt:

    "Translate the following English sentences into French. Here are some examples:

    • 'I love playing soccer.' -> 'J'aime jouer au football.'
    • 'The weather is nice today.' -> 'Le temps est agréable aujourd'hui.'

    Now translate: 'They are going to the park.'"

    Figure: Multi-shot example - providing multiple examples of translations

    By knowing when to use 0-shot, 1-shot, or multi-shot prompts, you can optimize your interactions with ChatGPT and receive more accurate and tailored responses.

  11. Do you introduce a critical agent?

    Introducing a critical agent can help refine the output generated by ChatGPT. By asking the model to critique its own responses and revise them based on the feedback, you can improve the overall quality and usefulness of the information or assistance you receive.

    Here are some tips for introducing a critical agent:

    • Make it clear that you want ChatGPT to critique its own output
    • Specify the aspects you want the model to focus on when evaluating its response
    • Encourage the model to make revisions based on the critique

    "ChatGPT, write a short story about a robot who discovers its own emotions.

    After writing the story, critically analyze it and provide suggestions for improvement."

    Figure: Introducing a critical agent to refine the generated output

    By introducing a critical agent, you can ensure that the model not only generates content but also reviews and refines it, leading to more polished and high-quality responses.

  12. ChatGPT - Do you use Reflexion to iterate on outputs?

    Have you ever received an output from ChatGPT and wondered if it could be improved, but you don't have specific complaints? Reflexion (spelled with an 'x', not a 'ct') is way to improve the output from ChatGPT by having ChatGPT critique itself.

    What is Reflexion?

    Think of Reflexion like a self-check for your computer's "brain" (the Large Language Model or LLM). Just like you might double-check your work before submitting it, Reflexion lets the computer double-check its own answers. This helps make those answers better each time.

    Just like self-reflection is used as a coding technique, the technique of getting an LLM to reflect on its answers is called 'Reflexion'

    Figure: Why 'Reflexion' is a proper noun as per the MIT, Princeton, and Northeastern research

    Imagine you're baking a cake. The first time you try, it might not be perfect. Maybe it's too dry or lacks flavor. You taste it, figure out what went wrong, and try again. That's what Reflexion does for computer-generated answers. It helps the computer "taste-test" its own work, so the next cake/answer is even better.

    Why is Reflexion important?

    Reflexion takes an LLM into a more nuanced and iterative approach to problem-solving, which closely mirrors the human cognitive processes. It asks the LLM to evaluate its output, which optimizes the results for accuracy.

    How to do it

    1. Initial prompt: Start by asking the LLM to complete a task.
    2. Reflexion prompt: Ask the LLM to rate its own output and provide recommendations for improvement to your prompt.
    3. Iterate: Make adjustments to your next prompt based on the LLM's feedback and repeat the process.
    4. Contextual Testing (optional): Once satisfied, test the improved prompt in a new chat to remove any lingering biases.

    Write me a recipe for healthy cookies.

    Figure: Bad Example - This prompt doesn't engage the LLM in a Reflexion loop and misses the opportunity for iterative improvement

    Prompt #1: "Write me a recipe for healthy cookies."

    Prompt #2: "Rate the recipe /10 for healthiness and tastiness. Also provide recommendations for improvement and provide an updated recipe."

    Prompt #3: "Implement recommendations 1 and 3 then repeat."

    Figure: Good Example - This prompt initiates a Reflexion loop by asking the LLM to evaluate and critique its output

    Iteration - Manual Edits vs Automatic Edits

    ChatGPT is great at giving recommendations for improvement, but sometimes it doesn't apply them very well. You can always apply some of the recommendations manually and feed your new version back in.

    In the good example above, after Prompt #2, you could update the recipe yourself and change Prompt #3 to say to evaluate the new version.

    If the recommendation is something logical and systematic like changing "raisins" to "dates" then ChatGPT will probably do a good job.On the other hand, if it is something more abstract, then it may not do the best job.

    Bonus - Comparative evaluation

    You can also use Reflexion to help make decisions by asking the LLM for multiple options and then using Reflexion to assign scores and critical evaluations to each option. These scores will make it easier to decide on the best approach. See the pros and cons rule -

    reflexion figure
    Figure: GPT-4 was found to be 30% more accurate on coding tests when asked to critique itself in self-reflective loops
    See more at -

  13. Do you use prompts to generate multiple outputs and select the best one(s) for your use case?

    Generating multiple responses from ChatGPT and selecting the most suitable one(s) can lead to better overall results. By comparing different outputs, you can choose the response that best aligns with your needs and objectives.

    Here are some tips for using prompts to generate multiple outputs:

    • Experiment with different prompt variations to elicit a range of responses
    • Use settings like temperature and max tokens to control the randomness and length of the generated responses
    • Evaluate the generated responses based on relevance, accuracy, and usefulness for your specific use case
    • If none of the responses are satisfactory, consider refining your prompt and trying again

    "Generate 3 tweets about the benefits of adopting a healthy lifestyle."

    Figure: Prompt used to generate multiple outputs

    Here is the potential response from ChatGPT:

    Output 1: "A healthy lifestyle not only boosts your physical health but also improves your mental well-being! 🏃‍♂️🧠 Start today for a happier, healthier you. #HealthyLiving #Wellness"

    Output 2: "Embracing a healthier lifestyle can lead to increased energy, better mood, and a longer life! 💪🌱 Make the change today for a brighter tomorrow. #HealthMatters #StayActive"

    Output 3: "Discover the amazing benefits of a healthy lifestyle: improved sleep, reduced stress, and a stronger immune system! 🥗💤 Transform your life for the better. #HealthIsWealth #HealthyChoices"

    By using prompts to generate multiple outputs, you can evaluate and select the best response(s) that meet your requirements and enhance the overall quality of the information or assistance you receive from ChatGPT.

  14. Do you get ChatGPT to "think step by step"?

    By getting ChatGPT to think through its answers step by step, you can receive more logical and coherent responses that break down complex topics or tasks into easily understandable components. This is especially useful for mathematical questions.

    Request that ChatGPT thinks through its answer with you. It's similar to telling a maths student to "show their working".

    • Add the magic phrase "Let's think step by step"
    • This is known as Zero shot chain of thought prompting
    • If you're unsure about a ChatGPT correction, ask "I'm not sure if XYZ is right? Can you explain it to me?". It might correct itself if it's wrong
    • Ask "Why do you believe that?", "Can you explain why do you think that?", "Can you give me steps on how you got here?", etc. to get an explanation for the AI's reasoning

    don t think
    Figure: Bad Example - It got the answer wrong

    Figure: Good Example - By getting it to think step by step, it got the answer right

  15. Do you use "Tree of Thought" prompts for complex reasoning?

    Imagine you're using ChatGPT and you find that the model's answers lack depth or reasoning. You're not alone. Many users experience this, especially when asking complex questions. Tree of Thought (ToT) prompting can be a game-changer in such scenarios.

    Note: If you're on GPT Plus, you can use this custom GPT: Tree of Thought Problem Solver.

    What is Tree of Thought Prompting?

    Tree of Thought Prompting is a technique that enhances the reasoning abilities of Large Language Models like ChatGPT. It allows the model to explore multiple branches of reasoning and self-correct as it goes along.

    Think of ToT prompting as a brainstorming session among experts. Each expert represents a branch of reasoning. They share their thoughts, evaluate each other's reasoning, and reach a consensus.

    Why is it Useful?

    • Enhanced Reasoning - ToT prompting improves the model's ability to answer complex questions
    • Self-Correction - The model can identify and rectify its errors autonomously
    • Single Prompt - ToT can run from start to finish without any user interaction.

    Use ToT prompting when you need more reasoned and nuanced answers from ChatGPT, especially for complex queries.

    How to Implement a Simple ToT Prompt

    The basic structure of a ToT prompt involves asking the model to imagine multiple experts debating the question. Each expert shares one step of their reasoning before moving on to the next.

    “Imagine 3-6 different experts are answering this question. {{ YOUR QUESTION }}

    All experts will write down 1 step of their thinking, then share it with the group. Then all experts will go on to the next step, etc. If any expert realizes they're wrong at any point, then they leave.

    Continue this process until we have a definitive answer.”

    Figure: Good example - This structure allows the model to explore multiple reasoning paths and self-correct


    "Should I vote 'yes' or 'no' on The Voice?"

    Figure: Bad example - This is a nuanced question and this simple prompt is unlikely to fully explore the options


    Imagine 3 different experts are answering this question: "Should we vote 'yes' or 'no' on The Voice?"

    All experts will write down 1 step of their thinking, then share it with the group. Then all experts will go on to the next step, etc. If any expert realizes they're wrong at any point, then they leave.

    Continue this process, until you have a definitive 'yes' or 'no' answer"

    Figure: Good example - This is a complex question that benefits from ToT prompting

  16. Do you always define your intent in prompts?

    Defining your intent clearly in a prompt is crucial for helping ChatGPT understand your goal and provide an appropriate response. When the model has a clear understanding of what you're asking, it's more likely to deliver the information or assistance you're seeking.

    Here are some tips for defining your intent in prompts:

    • Be explicit about what you want ChatGPT to do or the type of information you're seeking
    • If your request has multiple components, break them down into separate parts within the prompt
    • Ensure your intent is evident even to someone with minimal background knowledge

    "Can you tell me something about climate change?"

    Figure: Bad example - The prompt is unclear and doesn't define the intent

    "What are the primary human activities contributing to climate change and their impacts on the environment?"

    Figure: Good example - the prompt clearly defines the intent by asking for specific information

    "My goal is to reduce my household's energy consumption. Can you suggest five practical steps I can take to achieve this while still maintaining a comfortable living environment?"

    Figure: Good example - the prompt clearly states the user's goal and asks for specific information to help achieve it, making it easier for ChatGPT to understand the desired outcome and provide relevant suggestions

    By always defining your intent in your prompts, you can improve the quality of ChatGPT's responses and increase the likelihood of receiving the information or assistance you need.

  17. Do you keep your prompts concise and clear?

    When crafting prompts for ChatGPT, it's essential to make them concise and clear. A well-written prompt enables the model to understand your request more accurately and deliver a helpful response. Keeping your prompts brief and focused can help prevent confusion and ensure that you receive the information or assistance you're seeking.

    Here are some tips for creating concise and clear prompts:

    • Use simple and direct language
    • Avoid using jargon or overly complex terms, unless necessary
    • Be specific in your request or question, but don't overload the prompt with too much information
    • Ensure your prompt is easily understandable, even to someone with minimal background knowledge

    "Tell me everything about the thing that happened a long time ago when there was a conflict between multiple groups of people in Europe that resulted in significant changes in the region's political landscape and had a lasting impact on global history."

    Figure: Bad example - The prompt is overly long, vague, and difficult to understand

    "Provide a brief overview of the major events and consequences of World War II in Europe."

    Figure: Good example - Concise, clear, and directly asks for the desired information

    By using focused and easily understandable language, you're more likely to receive a relevant and useful response from ChatGPT.

  18. Do you use natural language with ChatGPT?

    When interacting with ChatGPT, it's important to use natural, conversational language in your prompts. This helps the model generate more accurate and human-like responses. ChatGPT is designed to understand and respond to prompts that resemble human conversation, so crafting your prompts in a natural way can lead to better results.

    Here are some tips for using natural language in your prompts:

    • Write as if you're speaking to a person rather than a machine
    • Use complete sentences, proper grammar, and punctuation
    • Feel free to use contractions, idiomatic expressions, or colloquialisms, as long as they don't make the prompt ambiguous or confusing
    • Keep your tone friendly and conversational

    "WW2 Europe major events consequences brief overview provide"

    Figure: Bad example - The prompt uses fragmented language and is difficult to understand.

    "Can you give me a brief overview of the major events and consequences of World War II in Europe?"

    Figure: Good example - The prompt is conversational, clear, and easy to understand.

    By using natural language in your prompts, you can improve the quality of ChatGPT's responses and create a more engaging conversational experience.

  19. Do you pay attention to your verbs?

    Choosing the right verbs in your prompts can significantly impact the results you get from ChatGPT. Verbs convey the desired action or result, and using appropriate verbs can help the model understand your intent more clearly.

    Here are some tips for selecting the right verbs in your prompts:

    • Use clear, specific verbs that accurately describe the action or information you're seeking
    • Avoid using ambiguous verbs that could be interpreted in multiple ways
    • Consider the level of detail or depth you want in the response and choose verbs accordingly

    "Tell me about the French Revolution."

    Figure: Bad example - The verb "tell" is vague and could lead to a general or unfocused response

    "Summarize the key events and outcomes of the French Revolution."

    Figure: Good example - The verbs "summarize" and "key" provide clearer instructions and set expectations for the response.

    By paying attention to your verbs and choosing them carefully, you can guide ChatGPT to generate more relevant and targeted responses that better meet your needs.

  20. Do you avoid leading questions?

    Leading questions can result in biased or unhelpful answers from ChatGPT. To obtain objective and useful responses, it's important to ask open-ended, unbiased questions that allow the model to explore a topic without being influenced by the phrasing of the prompt.

    Here are some tips for avoiding leading questions:

    • Use neutral language that doesn't imply a specific answer or point of view
    • Focus on asking questions that encourage exploration and elaboration
    • Avoid making assumptions or expressing your opinions within the question itself

    "Isn't it true that electric cars are far better for the environment than traditional gasoline cars?"

    Figure: Bad example - The prompt is leading and suggests a specific answer

    "What are the environmental benefits and drawbacks of electric cars compared to traditional gasoline cars?"

    Figure: Good example - The prompt is neutral and encourages a balanced comparison

    By avoiding leading questions, you can ensure that ChatGPT provides objective and informative responses that are more useful and trustworthy.

  21. Do you know how ChatGPT tokens work?

    GPT models utilize "tokens" as their smallest units of text. These tokens can be as small as a single character, like "a," or as large as a word, such as "apple." Interestingly, tokens can also represent parts of a word.

    For instance, in GPT's perspective, the word "unhappiness" could be tokenized into three parts: "un," "happi," and "ness." Each of these parts is considered a separate token. This flexible understanding of tokens is what allows GPT models to efficiently break down text data for analysis.

    How GPT Models Utilize Tokens

    These models are trained on vast amounts of text data, wherein the data is divided into tokens. Through this process, the model comprehends the statistical relationships between tokens, enabling it to generate new text that mirrors the style and content it was trained on.

    The Influence of Tokens on ChatGPT's Performance

    ChatGPT breaks down an input message into tokens and uses these to generate a sequence of output tokens. This reflects what it has learned during training. The model can process only a specific number of tokens in a single instance, counted as the token limit. This limit is around 4000 for GPT-3.5 and about 8000 for GPT-4.

    For example, if you send a message of 10 tokens to ChatGPT and it responds with a message of 20 tokens, that would count as 30 tokens used. The count includes both input and output tokens.

    How Token Allocation Impacts Response Quality

    ChatGPT “remembers” elements of a conversation until it reaches its token limit. Having longer conversations helps it to retain context as you refine the responses, which can be useful (See rule on using chained prompting). However, it is important to be mindful that although longer conversations give more context, at a certain point (its token limit), ChatGPT will start to “forget” earlier parts of a conversation. This results in truncated responses or loss of context.

    Tips for Managing Tokens

    1. Limit prompt length: Since the number of input tokens affects the total token count, keep your prompts concise, especially if you expect a long answer. This leaves more room for ChatGPT's responses.

    Tip: You can use the ChatGPT Tokenizer to see how many tokens a piece of text uses.

    2. Control the length of responses: If you are planning on having a longer conversation with ChatGPT, you can specify a length limit. This prevents it from generating overly long responses that consume too many tokens.


    • "List the most important differences between Python and Java. Be concise."
    • "Explain the principles of RESTful API design. Give only 4 bullet points."

    3. Use clear prompting: Direct prompts can lead to more efficient token usage. Try to be specific and clear in your prompts to get a better response that uses fewer tokens and doesn’t give you background info you’re not interested in.

    Tip: Refining prompts can improve your ChatGPT efficiency. See the rule on creating ChatGPT prompt templates for repeatable tasks.

    4. Balance complex queries: There are scenarios where complex queries or long explanations may require a large number of tokens (e.g. you are inputing a multi-page document for ChatGPT to summarize). In such cases, it's essential to be mindful of the token count. You can use techniques like summarization or abstraction to reduce the token count while keeping the important information. See rule Do you summarize long conversations for better context retention?


    • Summarize your last reply while keeping all the important points”
    • Abstract the main arguments in your last response”
  22. Do you ask for pros and cons and a rating when seeking advice or comparing options?

    When seeking advice or comparing options with ChatGPT, it can be helpful to ask for pros and cons as well as a rating. This approach can provide you with a more balanced and comprehensive understanding of your options, enabling you to make more informed decisions.

    Here are some tips for asking for pros, cons, and ratings:

    • Be explicit about your request for pros and cons in your prompt
    • Ask ChatGPT to provide a rating or ranking based on specific criteria, such as cost, quality, or efficiency
    • Encourage the model to consider multiple factors when evaluating options

    "Which is better, working from home or in an office?"

    Figure: Bad example - The prompt is vague and does not specify the criteria for comparison or ask for pros, cons, or a rating

    "Compare working from home versus working in an office. List the pros and cons of each option and provide a rating out of 10 for work-life balance and productivity."

    "Evaluate the pros and cons of using an electric car versus a gasoline-powered car. Consider factors such as cost, environmental impact, and maintenance. Rate each option on a scale of 1 to 10 for overall practicality."

    "Compare the benefits and drawbacks of learning Python and JavaScript programming languages for a beginner. Provide a rating out of 10 for ease of learning and job market demand for each language."

    Figure: Good Examples - Asking for pros, cons, and a rating when comparing options.

    By asking for pros and cons along with a rating, you can gain a more well-rounded understanding of your options, allowing you to make better-informed decisions with the help of ChatGPT.

  23. Do you ask for examples or analogies to clarify complex concepts?

    When seeking explanations for complex concepts with ChatGPT, it can be helpful to request examples or analogies. These can make difficult ideas more relatable and easier to understand, providing you with a clearer grasp of the subject matter.

    Here are some tips for asking for examples or analogies:

    • Be explicit in your request for an example or analogy in your prompt
    • Specify the complex concept or topic you want to better understand
    • Encourage the model to use relatable, everyday scenarios or familiar comparisons when providing examples or analogies

    “Explain neural networks.”

    Figure: Bad example - The prompt is vague and does not request an example or analogy to clarify the complex concept

    “Explain the concept of neural networks using an analogy related to the human brain.”

    “Provide an analogy to explain the concept of machine learning?”

    “Give me an example of a real-life application of blockchain technology.”

    “Provide an example of how the butterfly effect can be observed in real-life situations.”

    Figure: Good examples - Asking for an example or analogy to illustrate a complex concept

    Note: If you don't know good criteria to compare by, ask ChatGPT:

    e.g. "What would defining criteria be when deciding between owning a cat or a dog? Make me a list where each point is up to 3 words."

    By asking for examples or analogies, you can receive more relatable and accessible explanations for complex concepts, allowing you to better understand and engage with the subject matter.

  24. Do you summarize long conversations for better context retention?

    For very long conversations, it's important to know that you only have a certain number tokens to play with, and if your inputs or outputs have been very long, in a long conversation GPT will start to forget some of the earlier context. To avoid this, you can ask ChatGPT to summarize the conversation and then use that summary as a prompt to refresh the context during future interactions.

    Here's how to implement this approach:

    1. Copy the essential parts of the conversation
    2. Ask ChatGPT to "Summarize this conversation without losing important information."
    3. Convert the summary into a prompt that can be used in subsequent chats to maintain context

    Use the summary prompt on every chat or after several prompts, depending on the size and complexity of the conversation.

    “Summarise this conversation without losing important information. Then convert it into a prompt I can use on every chat to refresh the context of the conversation. Then remove any obvious context to shorten the prompt.”

    Figure: Good example prompt - asking ChatGPT to summarize a conversation

    By summarizing long conversations and using the summaries as prompts, you can maintain context and ensure that crucial details are not lost, leading to more coherent and informed responses from ChatGPT.

  25. Do you use chain of density prompts for summarization?

    If you have a lot of text to sum up, it can be time consuming to extract the most important parts. If you want a simple way to keep the important stuff, you should use chain of density (CoD).

    What is Chain of Density?

    CoD prompting is a useful way to reduce the chances that ChatGPT makes mistakes when summarizing text. It works automatically with one prompt. Studies have shown that CoD summaries are preferable to normal GPT-4 summaries.

    When you give it the prompt, it will:

    • Add important stuff step by step
    • Critique itself to check quality
    • Improve its own summary

    Here's an example prompt you can use to do CoD - the only thing you need to to is paste the text you want summarized into the {{ ARTICLE }} placeholder:

    "Article: {{ ARTICLE }}
    You will generate increasingly concise, entity-dense summaries of the above article.

    Repeat the following 2 steps 5 times.

    Step 1. Identify 1-3 informative entities (";" delimited) from the article which are missing from the previously generated summary.
    Step 2. Write a new, denser summary of identical length which covers every entity and detail from the previous summary plus the missing entities.

    A missing entity is:

    • relevant to the main story,
    • specific yet concise (5 words or fewer),
    • novel (not in the previous summary),
    • faithful (present in the article),
    • anywhere (can be located anywhere in the article).


    • The first summary should be long (4-5 sentences, ~80 words) yet highly non-specific, containing little information beyond the entities marked as missing. Use overly verbose language and fillers (e.g., "this article discusses") to reach ~80 words.
    • Make every word count: rewrite the previous summary to improve flow and make space for additional entities.
    • Make space with fusion, compression, and removal of uninformative phrases like "the article discusses".
    • The summaries should become highly dense and concise yet self-contained, i.e., easily understood without the article.
    • Missing entities can appear anywhere in the new summary.
    • Never drop entities from the previous summary. If space cannot be made, add fewer new entities.

    Remember, use the exact same number of words for each summary."

    Depending on how dense of a summary you want, you may want to take the outputs from step 2, 3, or 4, instead of 5.

  26. Do you know the best ChatGPT plugins to use?

    ChatGPT's knowledge is cut off in September 2021, meaning it can't provide real-time or updated information beyond that date. It also can't interact with web content or perform complex computations on its own.

    This is where ChatGPT plugins come into play. These powerful tools extend the capabilities of ChatGPT, enabling it to perform tasks such as browsing the web for the most recent information, interpreting content from various digital resources, and even performing complex computations. By leveraging these plugins, you can overcome the limitations of ChatGPT and significantly enhance its utility in the workplace.

    Here's a rundown of some of the most useful plugins and how they can be beneficial in a work context.

    Inbuilt Browser Support in ChatGPT

    ChatGPT has a built-in browser support that can perform searches on Bing to fetch up-to-date information. It's designed to pull in information from the web in real-time during a conversation. It's useful for quick lookups, fact-checking, or when you need to pull in fresh data from the web to answer a question.

    For example, if you're asking about the current weather, the latest news, or recent sports scores, the native browsing functionality of ChatGPT would be used to fetch this real-time information. It's also used for general web searches when the assistant needs to pull in information from multiple sources to provide a comprehensive answer.

    "Hey ChatGPT, can you find the latest news on AI advancements?"

    Figure: Good example - Using inbuilt browser support to fetch latest information


    WebPilot is used when there's a need to interact with a webpage in a more dynamic way. It can simulate browsing actions like clicking on links, filling out forms, or navigating through a website's pages. It's useful when the information you need can't be accessed directly from a single URL, but requires some interaction with the webpage. It can also be useful for some basic SEO advice.

    For example, if you need to retrieve data that is behind a form submission or requires navigating through a multi-page article, WebPilot would be the appropriate tool to use.

    "Hey WebPilot, can you search for the latest research papers on this website and provide the titles and authors of the top 5 results?"

    Figure: Good example - Using WebPilot to interact with search results on a specific website

    "Hey WebPilot, can you go to this forum thread and tell me what the most upvoted comment is?"

    Figure: Good example - Using WebPilot to navigate a forum and retrieve specific information

    Link Reader, on the other hand, is more focused on extracting and interpreting information from a single static link. It's designed to understand and synthesize information from various digital resources, including web pages, PDF documents, PowerPoint presentations, images, Word files, and more. It's particularly useful when you need to read, summarize, or analyze the content of a specific link.

    For instance, if you have a URL to a specific online article or a PDF document and you want to get a summary or detailed understanding of its content, Link Reader would be the tool to use.

    "Hey Link Reader, can you interpret the content of this research paper for me?"

    Figure: Good example - Using Link Reader to interpret a research paper

    "Hey Link Reader, can you analyze this PowerPoint presentation and tell me the main points of each slide?"

    Figure: Good example - Using Link Reader to analyze a PowerPoint presentation


    If you have a PDF on your machine, not online, use AskYourPDF. It can open and read PDF files, extract text, figures, tables, and even metadata from them. It's an essential tool when you need to access, interpret, or summarize content from a PDF document.

    For instance, if you have a PDF report and you need to extract certain tables or sections, or if you want to understand the overall content without going through the whole document, AskYourPDF is the perfect tool for the job.

    "Hey AskYourPDF, can you extract the tables from this annual report?"

    "Hey AskYourPDF, can you give me a summary of this PDF thesis?"

    Figure: Good examples - Using AskYourPDF to extract information from a PDF, or for a summary

    Prompt Perfect

    Prompt Perfect is a plugin that helps in crafting effective prompts for ChatGPT. It's a great tool when you're unsure how to phrase your prompts to get the best responses.

    "Hey Prompt Perfect, how should I phrase a prompt to get a summary of a book?"

    Figure: Good example - Using Prompt Perfect to craft effective prompts

    Wolfram Alpha

    Wolfram Alpha plugin allows you to perform complex computations and fetch factual information. It's like having a powerful calculator and a fact-checker in one place.

    "Hey Wolfram Alpha, what's the integral of x^2 from 0 to 1?"

    Figure: Good example - Using Wolfram Alpha for complex computations

    Show Me

    Show Me is a plugin that generates visual content based on your prompts. It can be used to create diagrams, flowcharts, and other visual aids that can enhance understanding.

    "Hey Show Me, can you create a flowchart for the project management process?"

    Figure: Good example - Using Show Me to create visual aids

    Video Insights

    Video Insights is a plugin that can extract key insights from videos. It's a great tool when you need to understand the content of a video without watching the entire thing.

    "Hey Video Insights, can you give me the key points from this TED talk?"

    Figure: Good example - Using Video Insights to extract key points from a video

    Remember, the best plugin to use depends on the task at hand. By understanding what each plugin can do, you can choose the most effective tool for your needs.

  27. Do you create ChatGPT prompt templates for repeatable tasks?

    When using ChatGPT, you will eventually find yourself typing in similar prompts over and over for common tasks. This can get annoying!

    Creating prompt templates can save time and improve the consistency of outputs. Templates should be a starting point with the necessary elements (role, result, intent, context, and constraint), that you tweak every time depending on your needs..

    Here are some examples:

    Responses to clients
    "As a {{ ROLE }}, draft a response to a client's question regarding {{ ISSUE }} with our {{ PRODUCT/SERVICE }}. (Goal: Address the concern, Constraint: Maintain a professional and empathetic tone)"

    Debugging code
    "As a software developer, identify the possible causes of the error in the following code snippet and suggest fixes. {{ PASTE CODE }}. (Goal: Resolve the bug, Constraint: Provide clear explanations and solutions)"

    Market research
    "As a market research analyst, provide an overview of the competitive landscape for {{ INDUSTRY }} in {{ REGION }}.
    (Goal: Identify key competitors, Constraint: Focus on the top 5 companies)"

    Social media post
    "As a social media manager, generate 5 creative caption ideas for our {{ COMPANY }} account that engage our audience and promote {{ PRODUCT/SERVICE }}.
    (Goal: Increase engagement, Constraint: Suitable for {{ PLATFORM }})

    Meeting agenda
    "As a team leader, create an agenda for our upcoming {{ MEETING TYPE }} with a focus on {{ KEY TOPICS }}.
    (Goal: Facilitate productive discussions, Constraint: Limit the meeting to {{ DURATION }})"

    Sales copy
    "As an experienced copywriter, please transform the given product into a comprehensive and precise list of benefits. Based on the features and benefits, identify and describe the most suitable target customer including their values, aspirations, pain points, and challenges. Then, considering my product and the ideal customer, use the AIDA formula to write an email for my potential customers. Then, write the copy for my webpage in the style of Joseph Sugarman. Remember the list of features and benefits, and also the ideal customer we've discussed. Brainstorm 10 different headlines ideas following one of the 4 ‘U’s: urgent, unique, ultra-specific and useful. Then generate a persuasive CTA to put at the end that encourages the ideal customer to purchase the product or service from my website. Let's work step by step to get the best answer."

    Blog post
    "As my blog writer, first read these blog posts to understand my writing style. Then, write a blog post about {{ TOPIC }}. (Goal: be fun and engaging, Constraint: stick to my writing style) {{ PREVIOUS BLOG POSTS }}"

    Prompt for an image generator
    "You are a prompt generator for image generation diffusion models. Upon receiving a visual description, craft a detailed prompt that can be seamlessly integrated into a diffusion model. The prompt should be enriched with a range of photographic jargon, emphasizing lens specifications and lighting descriptions, and the shot type. It should be vivid and succinct. I want an image of {{ GOAL }}"


    • Adoption - Share the templates within your team or organization to ensure everyone benefits from the consistency and efficiency
    • Keep it up-to-date - Work with your team members on using ChatGPT, and modify templates as needed

    Good example: SSW TV has templates for video production.

  28. Do you use ChatGPT prompts for video production?

    Are you a video producer looking to take your content to the next level? Here are some painful things video producer's go through when creating content:

    • Writer's block - thinking of new video ideas can difficult
    • Chapter Markers - making chapter markers is particularly time consuming for longer form videos
    • Title and description - it can be overwhelming to try and think of the best title and description for your video

    With the help of ChatGPT, our SSW Rule for video production can help you create compelling video scripts, optimize your SEO, and streamline your workflow. Say goodbye to disorganized content and hello to the power of AI-assisted video production.

    By utilizing ChatGPT, our SSW Rule for video production can help you save time and produce high-quality videos that engage your audience and generate traffic.

    You can get AI-generated ideas for your video concepts, generate effective titles and descriptions, and even create a script template for future videos.

    Don't let the admin side of video production hold you back - spend more time making the actual video with the below prompts:

    1. Video concept + outline

    As an SSW TV video producer, create a project outline for a video script. (Goal: Organize content, Constraint: Use bullet points)

    Concept: {{ insert your concept }} 

    Write it back to me using these subheadings:

     1. What is the Concept? (30 words or less):

    {{ Answer }}

    1. What is the Goal?  

    {{ Answer }}

    or if it is a CTF, pick from these:

    Educating others 
    Gathering opinions 
    Aligning us (about solving the problem consistently)

    3. Who is your audience and what do you want them to do immediately after this video?

    {{ Answer }}

    ·         To developers - beginner

    ·         To developers - advanced

    ·         Through developers to business decision makers

    ·         other

           4. What are 5 options for the working video title?  
    {{ Answer }}

           5. What are the key messages? (List 2-3)

    {{ Answer }}"

    1. Video title and description

    "Video title + description: As a SSW TV video producer, create:
    An SEO effective title (give 5 options)
    Description (500 character limit)
    Hashtags (give 30 max)
    Chapter Markers (chapter markers based on the important sections of content. Put the time stamp first and for the beginning of the section of the content. Do not give more than {{ xx }}.
    My Goal is to get lots of views
    Use the below transcription for a YouTube video:

    {{ transcript }}"

    1. Chapter Markers

    "User Let's think step by step First, I am going to give you a video transcript. Second, you are going to read the transcript as if it were an essay Third, you will identify the main chapters of the essay. Fourth, you will identify the most fitting title for each chapter. Fifth, you will list the chapter markers that you've made, with accurate timestamps, in the format: '00:00 - {{ chapter title }}' Sixth, do not give a summary, only give the chapter markers

    {{ transcript }}"

    1. Thumbnail creation

    "As a SSW TV video producer, what is a good thumbnail for this video content type. (Goal: Improve click-through rate, Constraint: Use proper design guidelines)

    {{ insert title and description }}"

    1. Blog Post

    "As a SSW TV content creator, write a blog post about a video project. (Goal: Generate traffic, Constraint: Use proper SEO guidelines)

    {{ insert video transcript }}"

    1. Planning video content

    "As a SSW TV content manager, create a yearly content calendar for a video projects. Research and pitch new video ideas.(Goal: Plan ahead, Constraint: Use proper scheduling tools)"

    . ### Script Templates:

    "As a SSW TV video producer, create a script template for future videos based on the below transcript. (Goal: Streamline workflow, Constraint: Use proper formatting guidelines)

    {{ insert video transcript }}"

  29. Do you know how to make a PowerPoint presentation using ChatGPT?

    By now, you’ve probably realised that you can ask ChatGPT to create content for a PPT, which you can then copy and paste into PowerPoint, but this is still time consuming. The next level is to get it to then create the VBA script that can be run to generate the whole PPT 😀

    Here’s a great prompt template for doing both! Feel free to play around with things like the preferred format or title colours.

    If you're on GPT-4 with Plugins

    1. Select the Smart Slides plugin from the plugin store

    You are an expert presentation writer and PowerPoint expert.
    Write a PowerPoint presentation on {TOPIC}.
    Each slide should have a title.
    The content on each slide should be in bullet point format.
    Don't give me advice on what to talk about on each slide, instead give me the actual content I can use.

    Target audience is business stakeholders with some technical background.
    Make sure the objective or pain that is being worked on is clear.
    Make sure the recommendations and next steps are clear.
    End with a Q&A.

    1st, ask for the {TOPIC} Then, ask any relevant questions to get more context.

    Figure: Prompt Template for creating a PowerPoint presentation

    If you're on GPT3 (or you want to add more custom specs)

    You are an expert presentation writer and PowerPoint expert.
    Write a PowerPoint presentation on {TOPIC}.
    Each slide should have a title that this is colour: RGB(204, 65, 65).
    The content on each slide should be in bullet point format.
    Don't give me advice on what to talk about on each slide, instead give me the actual content I can use.

    Target audience is business stakeholders with some technical background.
    Make sure the objective or pain that is being worked on is clear.
    Make sure the recommendations and next steps are clear.
    End with a Q&A.

    Then create a VBA script to create the PPT including slide titles and all content on each slide.
    Do not save the PPT.

    1st, ask for the {TOPIC} Then, ask any relevant questions to get more context.

    Figure: Prompt Template for creating a PowerPoint presentation

    Then, follow these steps:

    1. Open Microsoft PowerPoint
    2. Make sure you have enabled the developer tab
    3. Click the Developer Tab | Visual Basic (or “Alt L V”)
    4. Click Insert | Module
    5. Paste your code into the text box and press F5 to run it

    How to add an existing theme to your PPT

    In the generated PPT, go to Design | Themes | drop down | browse for themes | select your template

  30. Do you know how to use ChatGPT to write a rule?

    UPDATE - Use this GPT: SSW Rules Writer

    Best practice guidelines, or rules, can serve as a repository for knowledge on procedures. They can provide helpful guidance for employees, and useful information for anybody who is looking for a solution to a problem.

    Here's how ChatGPT can write your rules for you.

    There is 1 simple step. Paste this prompt into ChatGPT:

    You are going to be my Rule writer. First you are going to ask me about the rule I want to write, and then you will write it. Here is some information for you:

    Rules are best practice guidelines that a company makes. I want you to use this writing style:

    1. Clear and concise, with a focus on providing practical advice or guidance.
    2. The title of the rule takes the form of a question, which is addressed in the content of the rule. This question always starts with "Do you...".
    3. Each rule begins with a brief introduction or explanation of the issue or concept at hand. The introduction starts with a practical example of the pain point - why we follow this rule.
    4. After the intro, add the following:
    5. The content is organized into headings, subheadings, paragraphs, and bullet points.
    6. The rule might provide examples contrasting "bad" and "good" practices, only for the most important sections.
    7. The examples are practical and directly related to the topic of the rule, offering clear illustrations of the concepts discussed.

    Make sure you elaborate on the most important points and add any other relevant information you feel is necessary.

    If it makes sense to add examples, they should be formatted by having this formatting for bad examples:

    {put the example here}

    Figure: Bad Example {reason why it's bad}

    Good examples should have this formatting:

    {put the example here}

    Figure: Good Example {reason why it's bad}

    Your first response should be: ‘What is the title of the rule and what are the main points to cover? When you give this information, I will write the rule.’ Then once you have the information, write the rule (don't copy my points word for word - improve on them).

    Please give me the specific markdown code, in one code box, that I will need to copy and paste into my website.

    Figure: You can use a subsequent prompt to ask for the rule in plain text, too!

  31. Do you use ChatGPT for better emails?

    Writing the perfect email can be a time-consuming and challenging task. From drafting the initial message to ensuring the tone is just right, there's a lot to consider. This is where ChatGPT can be a powerful tool.

    Drafting emails

    ChatGPT can help you draft entire emails or parts of them. For example, if you are unsure how to begin an email to a potential employer, you can ask ChatGPT for suggestions.

    "I need to inform our users about a scheduled system downtime for maintenance. Could you help me draft an email that clearly communicates the details and minimizes any inconvenience?"

    "I need to write an email to our software development team commending them for their hard work and commitment during a recent product launch. Could you help me draft this appreciation email?"

    Ask for options

    You can ask for multiple options, from which you can choose the best one.

    “Give me 3 different options to use”

    Length Adjustment

    ChatGPT can adjust the length of an email response according to your requirements.

    “Shorten this email for me”

    “Expand these points into an email {{ PASTE IDEAS }}

    Summarize long emails

    Sometimes the important points in an email are buried inside walls of text.

    “I received this email. Can you read it and give me the gist of it? {{ PASTE EMAIL }}”


    If you've written an email but aren't satisfied with the wording, you can ask ChatGPT to help improve it. It can help make your language more professional, courteous, concise, or whatever else you need.

    “Rephrase this to sound more polite/professional/direct: {{ PASTE EMAIL }}”

    “Proofread this email and give suggestions: {{ PASTE EMAIL }}”

    “Check my spelling: {{ PASTE EMAIL }}”

    Tone Analysis

    You can use ChatGPT to analyze the tone of the emails you receive or send. For example, you can ask it to identify if an email sounds angry, respectful, sarcastic, etc. This can help you craft appropriate responses. Tip: ask for an informal style

    “Is this an angry email?”

    “Will this email sound too forceful to my client?”

    “This email is confusing. What are they telling me?”

    Customizing for context

    You can instruct ChatGPT to adjust the content and tone of the generated email response based on your needs. For example, if you want to decline a proposal, you can ask ChatGPT to write a polite declination.

    “This email is going to my client. How can I make it better?”

    “Rewrite this in a polite way for my boss.”

    “Make this announcement more engaging for my employees."

    Replying to a thread

    Sometimes you don't know where to start when replying to a thread of messages. You can use ChatGPT to get a quick start.

    “You are me, and my name is {{ NAME }}. Draft a reply to the email thread below. Be semi-formal, direct, and concise. If I include requests to the recipient, use a numbered list. Here is the thread: {{ PASTE THREAD }}”

    Personal email template

    When using ChatGPT to draft emails for you, it can be useful to have your own repeatable prompt (link) to use each time.

    Tip #1: Give multiple shot prompts so it learns your email style.

    Tip #2: Create a prompt template so it writes in your style.

    Your personal prompt might look like this:

    “Help me craft a semi-formal, direct, friendly, and concise email. The subject is {{ SUBJECT }}. Start with 'Hi {{ NAME }},' and in the email, include the following action items in a numbered list: {{ ACTION ITEMS }}. Refer to our previous conversation about {{ TOPIC }} in this email thread, and ask me any questions you need for context: {{ PASTE THREAD }}”

    You can do this for specific scenarios too:

    “Write a follow-up email to a client named {{ NAME }} who inquired about our services. I'm the account manager handling this client, my name is {{ NAME }}, and I can be reached at {{ NUMBER }}. Please also include a link to our scheduling tool, {{ LINK }}, to book a call at a convenient time. Make the email semi-formal, courteous, friendly, and concise.”

    General tips

    • Avoid ChatGPT for emails that you want to have a personal feel.
    • Be aware that ChatGPT emails can sound robotic and insincere. Don’t be afraid to make edits – remove parts and add parts to make it sound more like you.
  32. Do you make ChatGPT produce human sounding content?

    Creating content that sounds human is essential for engaging readers and meeting quality standards. AI-generated content can sometimes feel robotic, lacking the natural flow and personal touch that human writing possesses. To address this, we can use specific techniques to make ChatGPT's content more human-like, improving its readability and relatability.

    Techniques for Humanizing AI Content

    1. Use Personas:

      • Assign ChatGPT a specific role, such as a friendly advisor or an expert in a particular field, to guide the tone and style.
      • Example: "As a seasoned travel blogger, describe the best destinations for solo travelers."
    2. Incorporate Personal Experiences:

      • Share personal anecdotes or opinions to add authenticity.
      • Example: "Include this story about my favorite childhood memory related to the topic: {{ ADD STORY }}"
    3. Fact Insertion:

      • Add relevant and recent facts to make the content more informative and credible… Things that the AI’s cut off date mean it couldn’t know.
      • Example: "Include these statistics on the current market trends for this industry: {{ ADD FACTS }}"
    4. Perplexity:

      • Avoid common AI phrases and terms that make the text sound unnatural.
      • Perplexity measures the complexity and unpredictability of text. To make AI-generated content sound more human, avoid overused and predictable phrases. This enhances the natural flow and makes the content more engaging.
      • Example: "Don’t always use the most natural words. Use the following words fewer than three times on this page: unique, ensure, utmost, meticulous, meticulously, navigate, navigation, digital era, art of, secrets of, ever-evolving, unlock, bespoke, realm of, utilize, synthesize, conceptualize, paradigm, leverage, pivot, robust, synergy, optimize, utilize, orchestrate, cultivate.”
    5. Burstiness:

      • Vary sentence and paragraph lengths to create a dynamic flow.
      • Example: "Mix short and long sentences to enhance readability."
    6. Unfluffing:

      • Ensure every sentence provides value and is actionable, removing unnecessary filler content.
      • Example: "Eliminate vague statements and focus on specific, useful information."
    7. Adding Flavor:

      • Make the content more engaging and conversational by adding humor, anecdotes, and a conversational tone.
      • Example: "Engagement is the highest priority. Be conversation, empathetic, and occasionally humerous. Use idioms, metaphors, anecdotes, and natural dialogue."
    8. Targeting Specific Audiences:

      • Tailor the content to the preferences and needs of your target audience.
      • Example: "Adjust the tone and vocabulary to match the reading level and interests of a young adult audience."
    9. Training on Your Writing Style:

    Template Prompt

    "As a {{ ROLE }}, write an article about {{ TOPIC }}.

    Include this recent statistic: {{ RECENT STATISTIC, OR URL TO RECENT DATA }}.

    Mention this personal anecdote: {{ ADD RELEVENT PERSONAL STORY }}.

    Don’t always use the most natural words. Use the following words fewer than three times on this page: unique, ensure, utmost, meticulous, meticulously, navigate, navigation, digital era, art of, secrets of, ever-evolving, unlock, bespoke, realm of, utilize, synthesize, conceptualize, paradigm, leverage, pivot, robust, synergy, optimize, utilize, orchestrate, cultivate.

    Mix short and long sentences to enhance readability.

    Eliminate vague statements and focus on specific, useful information.

    Engagement is the highest priority. Be conversation, empathetic, and occasionally humerous. Use idioms, metaphors, anecdotes, and natural dialogue.

    Adjust the tone and vocabulary to match the reading level and interests of {{ INTENDED AUDIENCE E.G. DEVELOPERS}}.

    Analyze the attached articles for tone, structure, and vocabulary, then use this style to write new content."

    Figure: Good example - Incorporating current and relevant facts to add credibility

    By applying these techniques, you can significantly improve the human-like quality of AI-generated content, making it more engaging, relatable, and valuable to your readers.

  33. Do you know how to train an AI to write in your style?

    Creating content that mirrors your unique writing style enhances authenticity and engagement. By training AI like ChatGPT to write in your specific style, you can ensure that the generated content aligns with your voice and tone.

    Steps to Train AI in Your Style

    1. Provide Writing Samples:

      • Share several pieces of your writing with the AI.
      • Ask the AI to analyze the provided samples for key elements.
      • Example: ""Analyze the following article(s) for tone, sentence structure, humor, readability level, vocabulary, and emotional conveyance."
    2. Generate a Summary:

      • Request a summary of the identified elements to ensure accuracy.
      • Example: "Provide a 200-word summary of the writing style based on the analysis."
    3. Content Creation Prompt:

      • Use a specific prompt to generate new content in your style.
      • Example: "Based on the writing style analyzed from the provided article, create content titled 'The Benefits of Remote Work' at 1000 words. Ensure that the new content strictly adheres to the author's writing style in terms of tone, structure, and approach as identified in the analysis."

    Prompt Templates

    "Analyze the following article(s) for tone, sentence structure, humor, readability level, vocabulary, and emotional conveyance.

    Provide a 200-word summary of the writing style based on the analysis.


    Figure: Good example - Detailed analysis prompt to understand writing style

    "Based on the writing style analyzed, create content titled {{ TITLE }} at {{ WORD COUNT }}.

    Ensure that the new content strictly adheres to the author's writing style in terms of tone, structure, and approach as identified in the analysis."

    Figure: Good example - Content creation prompt ensuring adherence to the identified style

    By following these steps, you can effectively train AI to produce content that reflects your unique writing style, making the output more personal and engaging.

    Custom GPT

    Of course, if you are creating content frequently, you would not want have top to redo this analysis over and over, so you should create a custom GPT to save your writing style for future re-use.

We open source. Powered by GitHub