The winner of Singapore’s 2023 GPT-4 Prompt Engineering competition, Shelia Teo, has shared a 47-page post demonstrating her mastery of AI techniques and her ability to apply them effectively in real-world scenarios. Studying and applying her CO-STAR framework and other methods has improved my prompts and gotten me better results in less time. Now, I want to highlight her work for you.

In the competition(and the article), Teo focused on four strategies for more effective prompting:

  1. [🔵] Structuring prompts using the CO-STAR framework
  2. [🔵] Sectioning prompts using delimiters
  3. [🔴] Creating system prompts with LLM guardrails
  4. [🔴] Analyzing datasets using only LLMs, without plugins or code —=

I’ve reviewed each section and highlighted ways to put her ideas to work.

Part 1: Structuring prompts using the CO-STAR framework

Imagine you’re at your desk, ready to use AI to complete some work. Where do you start to ensure the AI understands your needs and produces relevant content? Enter the CO-STAR framework, which breaks down your prompt into Context, Objective, Style, Tone, Audience, and Response. This method guides the AI to deliver tailored, high-quality content.

For a recent project,  using the CO-STAR framework helped me generate a focused and compelling grant proposal. By providing context about my nonprofit’s mission and the program we were seeking funding for, stating my objective to secure a grant, specifying a persuasive, credible style and empathetic tone, identifying the grant committee as my audience, and outlining the desired proposal structure, I prompted the AI to create a draft that was an excellent basis for further writing and editing.

Follow these steps to apply the CO-STAR framework to your nonprofit’s AI interactions. This will help you structure your prompts effectively and guide the AI to generate content that aligns with your goals. 

  1. Context: Begin by providing background information about your nonprofit and the specific project or task you need assistance with. The information could include your organization’s mission, values, and the purpose of the content you’re generating.
  2. Objective: Clearly state what you want the AI to accomplish. Be specific about the type of content you need, such as a grant proposal, press release, or social media post.
  3. Style: Specify the writing style you want the AI to use. Depending on your goals and the type of content you’re creating, this could be formal, persuasive, or storytelling.
  4. Tone: Define the tone you want the AI to adopt. Depending on your target audience and the purpose of the content, this might be empathetic, inspiring, or informative for nonprofits.
  5. Audience: Identify the content’s audience, such as donors, volunteers, beneficiaries, or the general public. This helps the AI tailor its language and approach to resonate with your target audience.
  6. Response: Outline the desired format for the AI’s response, including any specific requirements like word count, structure, or key elements to include.

Putting more care into structuring your prompts can make them more precise and efficient–and generate higher-quality, targeted results.

Part 2: Sectioning prompts using delimiters

To enhance clarity and organization in your prompts, Teo advocates using delimiters—special characters or tags that segment your prompt into manageable parts. This ensures the AI processes each section correctly, enhancing the response’s precision. With this level of organization, the AI can process and respond to even the most nuanced and multilayered prompts with ease and precision.  

To bring the power of delimiters to your nonprofit’s AI interactions, follow these steps: 

  1. Choose your delimiters: Select a set of unique characters (e.g., ###, ===, >>>) or XML-style tags (e.g., <section> and </section>) that won’t be confused with regular punctuation or content.
  2. Break down your prompt: Divide your already-written prompt into sections, such as background information, specific instructions, and desired output format. Use your delimiters to separate these sections.
  3. Provide clear instructions: Within each section, offer concise and clear instructions to the AI, ensuring it understands the context, objectives, and expectations for the task at hand.
  4. Be consistent: Use the same delimiters throughout your prompt to maintain clarity and structure.

For example, if you’re asking an AI to help draft a grant proposal, your prompt might look like this:

<context> Our nonprofit provides job training and placement services for individuals with disabilities. We are applying for a grant to expand our programs to serve more clients in our community. </context> 

<instructions> Please help draft a compelling grant proposal highlighting our organization’s mission, impact, and plans for program expansion. Include relevant data on the need for our services and the expected outcomes of the proposed project. </instructions> 

<format> The proposal should be at most 1,000 words and include the following sections: Executive Summary, Needs Statement, Program Description, Expected Outcomes, and Budget Summary. </format>

Structuring your prompts with delimiters can help the AI generate more accurate, targeted, and compelling content as a response to your query.

Part 3: Creating system prompts with LLM guardrails

Teo’s third strategy involves integrating system prompts with built-in guardrails. This approach helps define clear boundaries and guidelines for AI interactions, which is beneficial for complex or sensitive content. This is yet another type of structure to experiment with for a more complex prompt, such as the outline for a strategic plan. 

To create a system prompt and set up guardrails for your nonprofit’s AI interactions, follow these steps:

  1. Define your task: Begin by clearly outlining the purpose of the conversation and what you expect the AI to accomplish. Tasks could include generating content, answering questions, or drafting the framework for a strategic plan.
  2. Set output format: Specify how you want the AI to structure its responses, such as using bullet points, numbered lists, or specific file types. Defining format helps maintain consistency and ease of use across multiple interactions.
  3. Establish guardrails: Identify any boundaries or guidelines you want the AI to adhere to throughout the conversation. Stating guardrails might include avoiding sensitive topics, maintaining a specific tone or perspective, or keeping your nonprofit’s values and mission statement top of mind in the response.
  4. Combine 2-4 into a system prompt: Integrate your task definition, output format, and guardrails into a single, overarching set of instructions for the AI. This system prompt will be applied to all responses within the conversation.

Here’s an example of a system prompt for a nonprofit focused on education:

<system_prompt> You are an AI assistant helping a nonprofit organization that provides education support for underprivileged students. Throughout this conversation, please ensure that your responses:

  1. Align with our mission of promoting equal access to quality education and empowering students from disadvantaged backgrounds.
  2. Provide relevant, age-appropriate resources and recommendations for students, parents, and educators.
  3. Maintain a supportive, inclusive, and empathetic tone, avoiding any language perceived as discriminatory or insensitive.
  4. Format responses as concise paragraphs with clear action steps or resource links when appropriate. </system_prompt>

By implementing system prompts and guardrails, nonprofit professionals can ensure that their AI-generated content remains consistent, appropriate, and mission-focused. This is another way to save time and effort while producing quality results.

Part 4: Analyzing datasets using only LLMs, without plugins or code 

Finally, Teo emphasizes the power of large language models (LLMs) to analyze data without additional software or coding skills. This is a game-changer for nonprofits that need to extract insights from data but need more technical resources. 

By feeding your data into an LLM and defining your analysis goals, you can surface game-changing insights in minutes – no data science degree required.

Here’s an example of using AI to analyze donor data to identify critical patterns and trends that will inform our fundraising strategies and donor engagement approaches.

Step-by-Step Guide:

  1. Data Preparation: Organize and clean your donor data before prompting the AI. This might include compiling data into a single spreadsheet or database with clear labels for each column, such as donation amounts, dates, donor demographics, and engagement history (e.g., event attendance, volunteer history).
  2. Defining Analysis Goals: Clearly articulate what you aim to learn from the data. For instance, you may want to identify:
    • Which campaigns were most effective at attracting large donations?
    • Trends in giving based on donor age groups or geographic locations.
    • Seasonal patterns in donation behaviors.
  3. Crafting the LLM Prompt: Develop a detailed prompt instructing the AI on what analysis to perform. Here’s an example prompt:

Screenshot

  1. Input Handling: Input the cleaned donor data file into the AI tool. This step may vary depending on your tool, but generally, you can upload the file directly or input data through an API if the platform supports it.
  2. Interpreting AI Analysis: Once the AI processes your prompt and dataset, it will output its analysis. This might include textual summaries, tables, or graphs depicting trends and patterns.
  3. Actionable Insights: Review the AI-generated insights to draw practical conclusions. For example:
    • If donation peaks are in December, consider launching your major annual fundraising campaign in November.
    • Tailor communications to demographics that show higher engagement, focusing on personalized approaches for top donors.
    • Use insights on donor retention to develop targeted strategies for improving engagement with first-time donors.
  4. Iterate and Refine: Based on the initial results, you may need to refine your prompts or data inputs to get more specific or accurate insights. Continuous interaction with the AI tool can help fine-tune the analyses to meet your needs better.

Following these steps, nonprofit professionals can utilize LLMs to conduct sophisticated data analyses that inform strategic decisions. This approach saves time and resources and enhances the organization’s ability to make data-driven decisions without needing specialized technical skills.

Conclusion 

Sheila Teo’s techniques offer the potential for nonprofits looking to enhance their efficiency and effectiveness through AI. By structuring prompts with the CO-STAR framework, using delimiters for clarity, implementing system prompts with guardrails, and harnessing LLMs for data analysis, your nonprofit can leverage cutting-edge technology to achieve good results.

The winner of Singapore’s 2023 GPT-4 Prompt Engineering competition, Shelia Teo, has shared a 47-page post demonstrating her mastery of AI techniques and her ability to apply them effectively in real-world scenarios. Studying and applying her CO-STAR framework and other methods has improved my prompts and gotten me better results in less time. Now, I want to highlight her work for you.

In the competition(and the article), Teo focused on four strategies for more effective prompting:

  1. [🔵] Structuring prompts using the CO-STAR framework
  2. [🔵] Sectioning prompts using delimiters
  3. [🔴] Creating system prompts with LLM guardrails
  4. [🔴] Analyzing datasets using only LLMs, without plugins or code —=

I’ve reviewed each section and highlighted ways to put her ideas to work.

Part 1: Structuring prompts using the CO-STAR framework

Imagine you’re at your desk, ready to use AI to complete some work. Where do you start to ensure the AI understands your needs and produces relevant content? Enter the CO-STAR framework, which breaks down your prompt into Context, Objective, Style, Tone, Audience, and Response. This method guides the AI to deliver tailored, high-quality content.

For a recent project,  using the CO-STAR framework helped me generate a focused and compelling grant proposal. By providing context about my nonprofit’s mission and the program we were seeking funding for, stating my objective to secure a grant, specifying a persuasive, credible style and empathetic tone, identifying the grant committee as my audience, and outlining the desired proposal structure, I prompted the AI to create a draft that was an excellent basis for further writing and editing.

Follow these steps to apply the CO-STAR framework to your nonprofit’s AI interactions. This will help you structure your prompts effectively and guide the AI to generate content that aligns with your goals. 

  1. Context: Begin by providing background information about your nonprofit and the specific project or task you need assistance with. The information could include your organization’s mission, values, and the purpose of the content you’re generating.
  2. Objective: Clearly state what you want the AI to accomplish. Be specific about the type of content you need, such as a grant proposal, press release, or social media post.
  3. Style: Specify the writing style you want the AI to use. Depending on your goals and the type of content you’re creating, this could be formal, persuasive, or storytelling.
  4. Tone: Define the tone you want the AI to adopt. Depending on your target audience and the purpose of the content, this might be empathetic, inspiring, or informative for nonprofits.
  5. Audience: Identify the content’s audience, such as donors, volunteers, beneficiaries, or the general public. This helps the AI tailor its language and approach to resonate with your target audience.
  6. Response: Outline the desired format for the AI’s response, including any specific requirements like word count, structure, or key elements to include.

Putting more care into structuring your prompts can make them more precise and efficient–and generate higher-quality, targeted results.

Part 2: Sectioning prompts using delimiters

To enhance clarity and organization in your prompts, Teo advocates using delimiters—special characters or tags that segment your prompt into manageable parts. This ensures the AI processes each section correctly, enhancing the response’s precision. With this level of organization, the AI can process and respond to even the most nuanced and multilayered prompts with ease and precision.  

To bring the power of delimiters to your nonprofit’s AI interactions, follow these steps: 

  1. Choose your delimiters: Select a set of unique characters (e.g., ###, ===, >>>) or XML-style tags (e.g., <section> and </section>) that won’t be confused with regular punctuation or content.
  2. Break down your prompt: Divide your already-written prompt into sections, such as background information, specific instructions, and desired output format. Use your delimiters to separate these sections.
  3. Provide clear instructions: Within each section, offer concise and clear instructions to the AI, ensuring it understands the context, objectives, and expectations for the task at hand.
  4. Be consistent: Use the same delimiters throughout your prompt to maintain clarity and structure.

For example, if you’re asking an AI to help draft a grant proposal, your prompt might look like this:

<context> Our nonprofit provides job training and placement services for individuals with disabilities. We are applying for a grant to expand our programs to serve more clients in our community. </context> 

<instructions> Please help draft a compelling grant proposal highlighting our organization’s mission, impact, and plans for program expansion. Include relevant data on the need for our services and the expected outcomes of the proposed project. </instructions> 

<format> The proposal should be at most 1,000 words and include the following sections: Executive Summary, Needs Statement, Program Description, Expected Outcomes, and Budget Summary. </format>

Structuring your prompts with delimiters can help the AI generate more accurate, targeted, and compelling content as a response to your query.

Part 3: Creating system prompts with LLM guardrails

Teo’s third strategy involves integrating system prompts with built-in guardrails. This approach helps define clear boundaries and guidelines for AI interactions, which is beneficial for complex or sensitive content. This is yet another type of structure to experiment with for a more complex prompt, such as the outline for a strategic plan. 

To create a system prompt and set up guardrails for your nonprofit’s AI interactions, follow these steps:

  1. Define your task: Begin by clearly outlining the purpose of the conversation and what you expect the AI to accomplish. Tasks could include generating content, answering questions, or drafting the framework for a strategic plan.
  2. Set output format: Specify how you want the AI to structure its responses, such as using bullet points, numbered lists, or specific file types. Defining format helps maintain consistency and ease of use across multiple interactions.
  3. Establish guardrails: Identify any boundaries or guidelines you want the AI to adhere to throughout the conversation. Stating guardrails might include avoiding sensitive topics, maintaining a specific tone or perspective, or keeping your nonprofit’s values and mission statement top of mind in the response.
  4. Combine 2-4 into a system prompt: Integrate your task definition, output format, and guardrails into a single, overarching set of instructions for the AI. This system prompt will be applied to all responses within the conversation.

Here’s an example of a system prompt for a nonprofit focused on education:

<system_prompt> You are an AI assistant helping a nonprofit organization that provides education support for underprivileged students. Throughout this conversation, please ensure that your responses:

  1. Align with our mission of promoting equal access to quality education and empowering students from disadvantaged backgrounds.
  2. Provide relevant, age-appropriate resources and recommendations for students, parents, and educators.
  3. Maintain a supportive, inclusive, and empathetic tone, avoiding any language perceived as discriminatory or insensitive.
  4. Format responses as concise paragraphs with clear action steps or resource links when appropriate. </system_prompt>

By implementing system prompts and guardrails, nonprofit professionals can ensure that their AI-generated content remains consistent, appropriate, and mission-focused. This is another way to save time and effort while producing quality results.

Part 4: Analyzing datasets using only LLMs, without plugins or code 

Finally, Teo emphasizes the power of large language models (LLMs) to analyze data without additional software or coding skills. This is a game-changer for nonprofits that need to extract insights from data but need more technical resources. 

By feeding your data into an LLM and defining your analysis goals, you can surface game-changing insights in minutes – no data science degree required.

Here’s an example of using AI to analyze donor data to identify critical patterns and trends that will inform our fundraising strategies and donor engagement approaches.

Step-by-Step Guide:

  1. Data Preparation: Organize and clean your donor data before prompting the AI. This might include compiling data into a single spreadsheet or database with clear labels for each column, such as donation amounts, dates, donor demographics, and engagement history (e.g., event attendance, volunteer history).
  2. Defining Analysis Goals: Clearly articulate what you aim to learn from the data. For instance, you may want to identify:
    • Which campaigns were most effective at attracting large donations?
    • Trends in giving based on donor age groups or geographic locations.
    • Seasonal patterns in donation behaviors.
  3. Crafting the LLM Prompt: Develop a detailed prompt instructing the AI on what analysis to perform. Here’s an example prompt:

Screenshot

  1. Input Handling: Input the cleaned donor data file into the AI tool. This step may vary depending on your tool, but generally, you can upload the file directly or input data through an API if the platform supports it.
  2. Interpreting AI Analysis: Once the AI processes your prompt and dataset, it will output its analysis. This might include textual summaries, tables, or graphs depicting trends and patterns.
  3. Actionable Insights: Review the AI-generated insights to draw practical conclusions. For example:
    • If donation peaks are in December, consider launching your major annual fundraising campaign in November.
    • Tailor communications to demographics that show higher engagement, focusing on personalized approaches for top donors.
    • Use insights on donor retention to develop targeted strategies for improving engagement with first-time donors.
  4. Iterate and Refine: Based on the initial results, you may need to refine your prompts or data inputs to get more specific or accurate insights. Continuous interaction with the AI tool can help fine-tune the analyses to meet your needs better.

Following these steps, nonprofit professionals can utilize LLMs to conduct sophisticated data analyses that inform strategic decisions. This approach saves time and resources and enhances the organization’s ability to make data-driven decisions without needing specialized technical skills.

Conclusion 

Sheila Teo’s techniques offer the potential for nonprofits looking to enhance their efficiency and effectiveness through AI. By structuring prompts with the CO-STAR framework, using delimiters for clarity, implementing system prompts with guardrails, and harnessing LLMs for data analysis, your nonprofit can leverage cutting-edge technology to achieve good results.