Intermediate to Advanced

Prompting Frameworks

April 2024

Jeff Glading and ChatGPT4

Turbo

1
Sections:
Intro

Freeform text inputs versus JSON formatted inputs

Prompting Frameworks:

1. GRADE (Goal, Request, Action, Details, Example:)

2. RODES (Role, Objective, Details, Example, Sense Check)

3. Chain of Thought (CoT)

4. Zero-Shot and Few-Shots Learning

5. ReAct (Reason and Act)

6. Instruction Tuning

7. Interactive Prompts

8. TRACI (Task, Role, Audience, Create, Intent)

9. TRAACI (Task, Role, Analyze, Audience, Create, Intent)

10. Scaffolded Prompts

11. SMART (Specific, Measurable, Achievable, Relevant, Timebound)

12. Prompt Chaining

13. Contextual Prompting

14. Contrastive Prompts

15. Meta Prompts
16. Dynamic Prompting

17. Multimodal Prompts

18. Ethical Prompting

19. Hierarchical Prompting

20. Guided Imagery Prompts

21. Recursive Prompts

22. Adaptive Learning Prompts

23. Cross-Modal Prompts

2
Welcome to "Prompt Domination Intermediate": Mastering the Art of AI
Prompting
In the rapidly evolving world of artificial intelligence, the ability to craft effective prompts stands
as a cornerstone for leveraging generative AI models' full potential. "Prompt Domination
Intermediate" is designed for users familiar with the basics of AI prompting and ready to delve
deeper into advanced strategies that refine interaction and enhance output quality from AI
systems.

This guide explores various prompting frameworks and input styles, each tailored to optimize AI
performance across multiple applications. Whether you're a developer, a content creator, or a
business strategist, understanding these sophisticated tools will empower you to command AI
technology more effectively, ensuring that your prompts lead to precise, relevant, and impactful
AI-generated responses.

Frameworks and Techniques Explored:

From GRADE to RODES and Chain of Thought to Dynamic Prompting, this guide covers 23
comprehensive frameworks designed to address specific needs and goals in AI interaction. You'll
learn each framework's definitions and practical applications, helping you choose the right tool
for any task.

Input Styles like Free Form Text and JSON are dissected to highlight their best uses.
Understanding these formats is crucial as AI models determine how prompts are structured and
processed, impacting everything from ease of use to integration with other systems.

Strategic Prompting for Advanced Users:

Delve into how each prompting framework can be adapted to different scenarios, whether you
are seeking creativity with Guided Imagery Prompts or precision with Structured JSON
Prompts.
Learn to balance flexibility and formal structure to maximize the effectiveness of your
interactions with AI. Whether scripting prompts for an interactive chatbot or defining tasks for a
data analysis AI, the choice of input style—adaptable Free Form Text or meticulously organized
JSON—can make all the difference.
By the end of this guide, you will have a thorough understanding of various advanced prompting
techniques and practical knowledge on implementing them effectively in your projects. "Prompt
Domination Intermediate" is your step toward mastering the complex yet rewarding world of AI
prompting, paving the way for innovation and efficiency in your interactions with artificial
intelligence.

The 23 prompting techniques covered are defined as follows:

3
1. GRADE (Goal, Request, Action, Details, Example): This framework helps structure
prompts to be more goal-oriented and actionable. It guides the AI to understand the goal,
acknowledge the request, perform actions based on it, consider all relevant details, and learn
from examples.

2. RODES (Role, Objective, Details, Example, Sense Check): Like GRADE, RODES
enhances AI's capability to deliver precise and contextually relevant responses. It includes a final
"sense check" to ensure the response aligns with the initial prompt.

3. Chain of Thought (CoT): This technique prompts the AI to articulate its reasoning process
step by step. It helps elicit more thoughtful and detailed responses from the model, improving
accuracy and comprehensibility.
4. Zero-Shot and Few-Shots Learning: These techniques involve prompting models without
examples (zero-shot) or only a few Examples (Few-Shots) to guide the AI's output. They
demonstrate the model's ability to adapt to tasks without extensive retraining.

5. ReAct (Reason and Act): Introduced by Google, this technique prompts the AI to generate
verbal reasoning and task-specific actions simultaneously, mimicking human-like decision-
making processes.

6. Instruction Tuning: This approach involves fine-tuning a model on a dataset of prompts and
their desired responses to improve its ability to follow instructions directly and produce the
intended outcomes.

7. Interactive Prompts: Engaging AI in a back-and-forth interaction where the user and the AI
collaboratively refine the output through iterative prompting, enhancing the result based on
continuous feedback.

8. TRACI (Task, Role, Audience, Create, Intent): This framework helps design prompts by
considering the task, the AI's role, the intended audience, what the AI should Create, and the
underlying intent. This structure aims to tailor highly relevant and effective prompts for specific
interactions.

9. TRAACI (Task, Role, Analyze, Audience, Create, Intent): An extension of TRACI, this
framework adds an analysis component to the prompting process, enabling a more thorough
examination of the input or situation before the AI generates its response. This addition is handy
for prompts requiring deeper insight or data interpretation.

10. Scaffolded Prompts: This technique provides a series of incrementally complex or guiding
prompts that help the model generate more detailed or contextually appropriate responses. It is
beneficial in educational settings or complex problem-solving tasks.

11. SMART (Specific, Measurable, Achievable, Relevant, Timebound): This framework,

adapted from goalsetting theory, can be applied to prompt engineering to ensure that AI-

4
generated tasks are clearly defined, quantifiable, realistic, aligned with broader objectives, and
time-sensitive.

12. Prompt Chaining: This technique involves creating a sequence of prompts where the output
of one serves as the input for the next. It's beneficial for complex tasks requiring multiple steps
or reasoning stages.

13. Contextual Prompting: This approach focuses on incorporating a rich contextual

background into the prompts to guide the model's responses more accurately. It is valuable for
tasks requiring deep domain knowledge or specific situational awareness.
14. Contrastive Prompts: This method uses pairs of prompts that illustrate "do this" versus
"don't do this" to help models learn to distinguish between desired and undesired outcomes. It is
especially useful for training models on nuanced tasks or finetuning their responses.

15. Meta Prompts: These are prompts about prompt making, where the AI generates or
optimizes other prompts. This recursive approach can help refine prompting strategies or develop
new ones.

16. Dynamic Prompting: This iterative process involves adaptable prompts based on the
model's responses or the user's real-time feedback. It is ideal for interactive applications and can
refine the model's output.

17. Multimodal Prompts: These prompts leverage multiple types of data inputs (e.g., text,
images, sound) to guide AI models, enabling richer and more complex interactions, particularly
in creative or integrative tasks.

18. Ethical Prompting: This focus is on designing prompts that consider ethical guidelines,
cultural sensitivities, and fairness to ensure that the model's outputs do not perpetuate biases or
harmful stereotypes.
19. Hierarchical Prompting: This method hierarchically structures prompts, starting from more
general concepts and moving towards more specific details. This can be particularly effective for
complex decision-making tasks or training models for layered information.

20. Guided Imagery Prompts: Used primarily in creative fields, these prompts guide AI in
generating visual content by describing scenes or objects in vivid detail, encouraging the model
to "visualize" and create complex images or designs.

21. Recursive Prompts: Like meta prompts, these involve prompts that use the AI's previous
responses as a basis to generate further refined outputs. This technique helps deepen the model's
engagement with a topic or problem over multiple iterations.

22. Adaptive Learning Prompts: This technique adjusts the complexity or style of the prompts
based on the AI’s performance, aiming to continuously challenge and train the model in a way
tailored to its evolving capabilities.

5
23. CrossModal Prompts: These prompts are designed to work across different modalities, such
as transitioning from text-based inputs to audio outputs, facilitating versatile applications,
especially in multimedia content creation.

6
Freeform text inputs versus JSON formatted inputs
As we draw together the strands from the vast tapestry of prompting techniques presented in this
guide, a central narrative emerges: the balance between Free Form Text's artful flexibility and
JSON formatting's structural robustness is vital to the mastery of prompt engineering. This
closing chapter reflects upon these two pivotal styles and how they interweave with the diverse
frameworks we've explored, influencing the efficacy and applicability of AI prompts.

1. Free Form Text: The Artisan's Tool

Flexibility: In Interactive Prompts and Guided Imagery Prompts, Free Form Text is akin to an
artist’s brush, offering broad strokes that invite imagination and personalization. Its adaptability
lends itself to the dynamic and organic nature of creative dialogue and narrative construction.

Ease of Use: Form Text shines in frameworks such as Ethical Prompting and Hierarchical
Prompting, where a more philosophical or conceptual touch is required. This format is the
layman's gateway to utilizing advanced AI, stripping away complexity and inviting intuitive
interaction.

Limitations: Yet, when we enter the territories necessitating meticulous detail, such as with
TRAACI or RODES, Free Form Text's lack of inherent structure may blur the fine lines that
define precision and specificity, leading to potential ambiguities in the AI's operations.

2. JSON: The Engineer's Blueprint

Structure and Precision: JSON's orderly nature provides a blueprint for success in frameworks
like GRADE and RODES, where clear definitions and systematic processes are paramount. Like
a well-conceived architectural plan, JSON lays out explicit instructions that ensure every element
is placed with intention and processed accurately.

Scalability and Integration: JSON is the superior scaffolding for digitally constructing complex
AI interactions, such as those required in Chain of Thought or Prompt Chaining. Its compatibility
with APIs and automated systems ensures that the intricate machinery of AI reasoning is well-
supported and scalable.

Complexity: However, this same precision may become a hindrance in the hands of those
unfamiliar with the nuances of programming languages. In frameworks like SMART or
Contrastive Prompts, where simplicity could be a virtue, the intricate requirements of JSON
syntax could be seen as unnecessarily cumbersome.

Comparison Across Frameworks

JSON's capacity for detailed compartmentalization ensures each logical step is faithfully
executed in frameworks demanding logic and adaptability, such as Chain of Thought or Dynamic
Prompting. Conversely, the creative spontaneity required by Meta Prompts or Guided Imagery
Prompts resonates more harmoniously with the expressive nature of Free Form Text.

7
Conclusion:

Each prompting framework's operational context and unique objectives guide the choice between
Free Form Text and JSON. JSON is the workhorse for tasks requiring the highest precision and
integration. But for those who benefit from ease and creativity, Free Form Text offers a canvas
ripe for innovation.

As we look to the future, let us carry forward the lessons learned from each framework, applying
the Free Form Text and JSON formats not as competing forces but as complementary tools in our
AI toolkit. This synergy will empower us to harness the full spectrum of AI's capabilities,
pushing the boundaries of what's possible and paving the way for a future where AI and human
creativity merge into a greater whole.

8
Section 1: Mastering the GRADE Framework
The GRADE framework is an instrumental tool in prompt engineering, designed to refine the
clarity and focus of prompts used in generative AI applications. GRADE stands for Goal,
Request, Action, Details, and Example. It helps structure prompts to ensure that each component
is purposeful and clearly defined, enhancing the AI's ability to produce relevant and precise
outputs.

Understanding the GRADE Components

1. Goal: Define the prompt's overarching objective. What do you hope to achieve with the AI's
response?

2. Request: Specify what you are asking the AI to do. This should be a direct appeal that aligns
with the goal.
3. Action: Detail the AI's actions to fulfill the request. This often involves outlining a process or
a series of steps.

4. Details: Provide additional information necessary for the AI to understand the context or
specifics of the task.

5. Example: Include a sample prompt and expected output to guide the AI, demonstrating how
to execute the request correctly.

Examples of the GRADE Framework in Use

Example 1: Content Generation

Goal: Generate a blog post about the latest AI trends.

Request: Write an informative and engaging article.

Action: Research recent developments in AI, focusing on specific areas like machine learning,
natural language processing, and robotics.
Details: The article should be approximately 1000 words, suitable for a tech-savvy audience, and
include data-driven insights.

Example: "Here's a brief on recent advancements in AI: [insert concise overview]. Based on
this, produce a detailed article discussing each trend."

Suggested input style: JSON, due to the need for structured data like word count and topics,
which are easier to specify programmatically.

Example 2: Customer Support Automation

Role: Assist customers with troubleshooting common issues.

9
Request: Provide step-by-step solutions.

Action: Identify the customer’s issue from the query, use the database to find the best solution,
and guide the customer through the resolution process.

Details: Responses should be concise and user-friendly, using polite and professional language.

Example: "For a customer asking about resetting their password, guide them through the
security steps and link to the password reset page."

Suggested input style: Free Form Text, allowing more natural, conversational interaction that
might be more engaging in a customer service context.
Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

Best for scenarios where the prompt requires flexibility and a natural, conversational tone, such
as creative writing or casual dialogues.

Suitable for less complex scenarios where the structure is not heavily dependent on multiple,
layered data points.

When to Use JSON:

This is ideal for prompts requiring high levels of detail and precision, where each prompt aspect
must be programmatically accessible and distinctly parsed.

It is useful when integration with other systems (e.g., databases, content management systems) is
necessary, allowing for efficient processing of detailed attributes like word count, specific data
points, or structured sequences.

Conclusion:

The GRADE framework is a powerful tool in prompt engineering, particularly effective in

environments where clarity and specificity drive better AI performance. By systematically
breaking down the components of a prompt, GRADE facilitates a deeper alignment between the
AI's capabilities and the user's objectives, ensuring that each interaction is purpose-driven and
effective. Whether using Free Form Text or JSON, the choice of input style should complement
the specific needs of the task at hand, optimizing both the user experience and the AI's response
accuracy.

10
Section 2: Navigating the RODES Framework
The RODES framework is a strategic tool in prompt engineering designed to enhance the
precision and contextuality of AI responses. An acronym for Role, Objective, Details, Example,
and Sense Check, RODES provides a structured approach to crafting prompts that are clear,
specific, and aligned with the intended outcome and context.

Understanding the RODES Components

1. Role: Define the capacity in which the AI operates. This sets expectations about the AI’s
persona or the professional lens through which it views the task.

2. Objective: Clearly state what the AI is expected to achieve. This will direct the AI’s efforts
and align them with the end goal.

3. Details: Provide additional context or parameters that the AI needs to consider. These can
include constraints, preferred methodologies, or specific considerations relevant to the task.

4. Example: Provide one or more instances to illustrate how the AI should perform the task. This
serves as a model for the desired approach and outcome.

5. Sense Check: This step ensures that the AI’s responses remain on track and align with the
initial prompt and objectives. It is a reflection or review stage to confirm correctness and
relevance.

Examples of the RODES Framework in Use

Example: 1: Project Management Assistant

Role: Act as a virtual project manager.

Objective: To organize and prioritize project tasks effectively.

Details: Consider project deadlines, team member availability, and task urgency.
Example: "Given a list of tasks with deadlines and team member schedules, generate a
prioritized task list for the week."

Sense Check: Verify that all high-priority tasks are scheduled early in the week and that every
team member is overbooked.

Suggested input style: JSON, due to the structured nature of the data (deadlines, schedules,
priorities) that needs to be programmatically parsed.

Example: 2: Customer Service Bot

Role: Served as a customer support representative for an online retailer.

Objective: Provide accurate and helpful responses to customer inquiries.

11
Details: Address customer queries regarding order status, product details, and return policies.

Example: Sense "If a customer asks about the return policy for an electronic item, guide them
through the process based on our standard return policy."

Sense Check: Ensure all provided information adheres to company policies and the responses
are clear and polite.

Suggested input style: Free-form text, which maintains a conversational and flexible interaction
style that can adapt to the varied nature of customer queries.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is ideal for applications where the AI must engage in natural, conversational interactions, such
as customer service or interactive storytelling.

It is best suited for scenarios where the prompt does not require rigid data structures or extensive
programmable logic.

When to Use JSON:

This is essential for tasks that involve complex data handling or when the prompts need to
interact with other digital systems, such as databases or project management tools.

This is preferred when the prompts must include multiple structured data points influencing the
AI’s task execution, such as analytics or detailed content generation.

Conclusion:

The RODES framework provides a comprehensive methodology for designing AI prompts that
are both effective and contextually relevant. By detailing the role, setting clear objectives,
providing rich contextual details, offering Examples, and incorporating a sense check, RODES
ensures that AI-generated responses are accurate, applicable, and aligned with the user's
expectations. Whether employing Free Form Text for more fluid, conversational prompts or
using JSON for structured, data-intensive scenarios, the choice of input style should strategically
complement the task's requirements, optimizing the interaction between the user and the AI.

12
Section 3: Exploring the Chain of Thought (CoT) Framework
The Chain of Thought (CoT) framework is a sophisticated technique in prompt engineering that
helps elicit detailed, logical reasoning from AI models. By encouraging the AI to process
information step by step and articulate its reasoning, CoT prompts facilitate a deeper
understanding and more comprehensive responses. This approach is beneficial in complex
decision-making tasks or when accuracy in reasoning is critical.

Understanding the CoT Components

The CoT framework involves structuring prompts that guide the AI through a logical sequence of
thoughts before concluding. This method helps generate more precise answers and aligns the
AI’s reasoning with humanlike thought processes.

Examples of the CoT Framework in Use

Example: 1: Solving Mathematical Problems

Prompt: "Imagine you are asked to solve the equation 2x + 5 = 15. Start by isolating the variable
x."

Chain of Thought: First, subtract five from both sides to get 2x = 10. Then, divide both sides by
2 to solve for x, which equals 5.

Suggested input style: Free-form Text allows the AI to express the reasoning process in a
natural, step-by-step narrative that is easy to follow.
Example 2: Business Strategy Development

Prompt: "A company wants to increase its market share. Outline a step-by-step reasoning
approach on how they might achieve this goal."

Chain of Thought: Begin by analyzing current market conditions and consumer behavior. Next,
identify potential areas for expansion or improvement in product offerings. Then, consider
strategic marketing campaigns that could attract more customers.

Suggested input style: JSON because it allows for structuring each thought process step as
separate elements that can be programmatically managed and analyzed, ensuring the response
maintains a logical flow.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best for educational or explanatory tasks where the AI needs to "think out loud" in a manner
that mimics human explanation, such as tutoring or customer support.

13
This is useful when the output needs to be engaging and easy to understand, making complex
ideas accessible through a conversational tone.

When to Use JSON:

It is ideal for tasks requiring integration with other systems that might process the AI’s output
further, such as in automated reasoning systems or complex analytical tools where each thought
process step might trigger different actions.

Effective when the reasoning process needs to be broken down into distinct steps that might be
evaluated or adjusted individually, enhancing the clarity and precision of the logic flow.
Conclusion:
The Chain of Thought framework significantly enhances the depth and quality of AI-generated
responses by promoting a structured, logical progression of ideas. Whether using Free Form Text
to foster a more natural and engaging narrative or employing JSON for its capacity to detail and
organize complex reasoning sequences, CoT can transform how AI models approach problem-
solving and decision-making. By tailoring the input style to the specific needs of the task,
developers and users can maximize the effectiveness of their AI systems, achieving outputs that
are not only correct but also contextually rich and logically sound.

14
Section 4: Zero-Shot and Few-Shots Learning Framework
Zero-shot and Few-Shots learning frameworks are transformative approaches in machine
learning that enhance the adaptability and efficiency of AI models. They enable them to perform
tasks without extensive training on large datasets. These frameworks are particularly valuable in
environments where data is scarce, or the model must be generalized from limited examples.

Understanding Zero-Shot and Few-Shots Learning

Zero-shot learning: In this approach, the AI is tasked with inferring correct responses without
having been explicitly trained on Examples of the specific task. It relies on the AI’s preexisting
knowledge and ability to generalize from related contexts.

Few-Shots Learning: This method involves training the AI on a minimal set of Examples or
data points specific to the task. It's used to quickly adapt the model to new tasks with minimal
input while maintaining high performance.

Examples of Zero-Shot and Few-Shots Learning Frameworks in Use

Example 1: Language Translation (Zero-Shot)

Prompt: "Translate the English phrase 'Good morning' into Japanese."

The model's Task is to use inherent linguistic knowledge to predict the correct translation without
direct training on this specific phrase.

Suggested input style: Free-form text allows the AI to apply its pre-trained language models
directly to the translation task without needing structured input.

Example: 2: Product Categorization (Few-Shots)

Prompt: "Given these examples, products and their categories, determine the category for a new
product described as 'Wireless Bluetooth Earphones.'"

Examples Provided: (1) "USB cable Electronics" (2) "Running shoes Footwear"
Suggested input style: JSON is ideal for structuring the examples clearly and allowing the
model to parse and process each product category pair efficiently.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is suitable for tasks where the AI must demonstrate comprehension or generate content based
on broad knowledge, such as answering questions, writing essays, or making predictions based
solely on preexisting training.
This is effective in scenarios where the prompt does not require detailed data parsing or when the
interaction is meant to mimic human-like conversations.

15
When to Use JSON:

This is ideal for scenarios where specific attributes or structured data points must be highlighted
within the prompt, especially in Few-Shots learning, where examples must be clearly defined
and distinguished.

Useful in applications that involve categorization, identification, or any task requiring the AI to
reference distinct metadata or properties.

Conclusion:

Zero-shot and Few-Shots learning frameworks are powerful tools that reduce the need for large
datasets and extensive retraining of AI models for new tasks. By leveraging these approaches,
users can effectively prompt AI systems to apply their preexisting knowledge or quickly adapt to
new information with minimal examples. The choice between Free Form Text and JSON
formatting will depend on the nature of the task at hand—whether it benefits from the flexibility
of unstructured prompts or requires the precision of structured data inputs. These methodologies
not only streamline the application of AI in diverse fields but also enhance the model's ability to
handle various tasks with impressive accuracy and efficiency.

16
Section 5: The ReAct (Reason and Act) Framework
The ReAct (Reason and Act) framework is an innovative approach in AI prompt engineering that
combines reasoning with action. This framework is particularly effective when an AI must
simulate a decision-making process akin to human thought and then execute a specific action
based on that reasoning. It is helpful for tasks requiring cognitive processing and subsequent
operational steps.

Understanding the React Components

ReAct prompts are designed to enable AI models to:

1. Reason: Articulate a logical thought process or reasoning leading to a decision or conclusion.

2. Act: Carry out an action based on the reasoning, which can be anything from generating text
to executing a function or modifying data.

Examples of the ReAct Framework in Use

Example 1: Recipe Adjustment

Prompt: "Explain how to adjust the sugar content in a cake recipe for a diabetic diet and then
modify the recipe accordingly."

Reason: The AI would discuss the effects of sugar on health, particularly for diabetics, and
suggest alternatives like using stevia instead of sugar.

Act: Provide a modified recipe with the exact amounts of stevia to use instead of sugar.
Suggested input style: JSON should clearly define the recipe's components and modifications in
a structured way that is easy for the AI to parse and modify.

Example: 2: Customer Support Scenario

Prompt: "A customer is asking for a refund because they received a damaged product. The
reason why they might be eligible for a refund and then draft an email response confirming the
refund process."

Reason: Analyze the company’s refund policy for damaged goods and assess the customer’s
situation against these criteria.

Act: Generate an email response to the customer confirming the refund and explaining the next
steps.

Suggested input style: Free-form Text for a more natural and conversational output appropriate
for customer communication.
Choosing Between Free Form Text and JSON Formatting

17
When to Use Free Form Text:

This is ideal for scenarios requiring detailed narrative responses where the AI must engage in a
more humanlike interaction, such as writing emails, reports, or customer service dialogues.

It is helpful in tasks where the reasoning needs to be expressed in a flowing, coherent manner
without the need for structured data input.

When to Use JSON:

It is best suited for tasks that involve data manipulation or require the AI to perform specific
actions based on structured data inputs, such as modifying database entries or adjusting
parameters in a technical setup.
This is essential in scenarios where each element of the reasoning and the corresponding action
needs to be distinctly captured and programmatically processed.

Conclusion:

The ReAct framework merges cognitive reasoning with practical execution, enabling AI systems
to understand and discuss tasks and perform necessary actions based on that understanding.
Whether employing free-form text for tasks requiring natural language generation or JSON for
more data-intensive operations, this framework enhances AI's capability to function
autonomously in complex environments. By implementing ReAct, developers can Create more
sophisticated, interactive, and practical AI applications that think and act efficiently in various
settings.

18
Section 6: The Instruction Tuning Framework
The Instruction Tuning framework is a sophisticated approach to prompt engineering that
involves finetuning an AI model to improve its understanding and execution of specific
instructions. This framework is beneficial for tailoring model responses to precise tasks,
enhancing the model’s performance in specialized applications without extensive retraining from
scratch.

Understanding the Instruction Tuning Components

Instruction Tuning typically involves:

1. Identification: Recognizing the specific instructions or types of tasks where the model's
performance could be improved.

2. Data Collection: Gathering examples that accurately reflect the instructions or tasks.

3. Model Training: Finetuning the model on this curated dataset to align its responses more
closely with the desired outcomes.

4. Evaluation: Testing the tuned model to ensure it performs better on the tasks.

Examples of the Instruction Tuning Framework in Use

Example 1: Legal Document Analysis

Task: "Identify and extract specific clauses related to data privacy from various contracts."

Data Collection: Gather a set of contracts containing diverse privacy clauses.

Model Training: Tune the model on this dataset to recognize and extract the relevant sections
accurately.

Evaluation: Assess the model's accuracy on unseen contracts to confirm improvement.

Suggested input style: JSON because it allows for the structured input of legal documents and
specific clauses that need to be identified, making it easier to finetune and test the model
systematically.

Example: 2: Customer Feedback Categorization

Task: "Classify customer reviews such as 'service', 'product quality', and 'user experience'."

Data Collection: Compile reviews that exemplify clear themes in each category.

Model Training: Adjust the model to improve its ability to categorize feedback based on these
examples.
Evaluation: Use separate reviews to evaluate the model’s classification accuracy.

19
Suggested input style: Free-form text, ideal for processing natural language inputs where the
categorization needs to capture the nuances of customer sentiment.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

Best for applications that require the model to process or generate natural language text, such as
customer service responses or content creation.

It is helpful in contexts where the tasks are defined by broad instructions rather than specific,
structured data points.
When to Use JSON:

It is essential for tasks that involve specific attributes or structured data, such as extracting
information from formatted documents or adjusting settings based on user input.

This is ideal for scenarios where each element of the instruction set can be clearly defined and
programmatically accessed, enhancing the precision of the tuning process.

Conclusion:

The Instruction Tuning framework enables AI models to perform specialized tasks more
effectively by realigning their understanding and execution capabilities toward clearly defined
objectives. Whether utilizing Free Form Text for more general language-based tasks or
employing JSON for structured, data-driven applications, Instruction Tuning is a powerful tool
for optimizing AI performance in targeted areas. By integrating this framework, developers can
ensure that their AI systems not only comprehend but also adeptly execute complex instructions,
thus broadening the practical applications of machine learning in industry-specific scenarios.

20
Section 7: The Interactive Prompts Framework
The Interactive Prompts framework facilitates dynamic and engaging dialogues between AI
models and users, where the conversation evolves based on input from both parties. This
approach enhances the AI's ability to understand and respond to user needs in real-time. It is
ideal for applications requiring adaptive responses, such as virtual assistants, interactive learning
environments, and customer service bots.

Understanding the Interactive Prompts Components

Interactive Prompts typically involve:

1. Initialization: This stage is set for interaction with an opening prompt that guides the user
through engagement.

2. Feedback Loop: Allowing the AI to ask questions and gather more information from the user
to refine its understanding and responses.

3. Adaptation: Dynamically adjusting the AI’s responses based on the user's inputs, enhancing
relevance and engagement.

4. Continuation: Maintaining the flow of conversation, encouraging ongoing interaction that

keeps the user engaged and informed.

Examples of the Interactive Prompts Framework in Use

Example 1: Virtual Shopping Assistant

Initialization: "Hi! I can help you find the perfect outfit. What type of clothing are you looking
for?"

Feedback Loop: The user responds, "I'm looking for a casual dress." The AI asks, "What color
and size do you prefer?"

Adaptation: Based on the user's preferences, the AI suggests a list of dresses that match the
described criteria.

Continuation: "Would you like to see more options or need assistance with something else?"

Suggested input style: Free Form Text to facilitate a natural and fluid conversation that mimics
human interaction.

Example 2: Interactive Language Tutor

Initialization: "Welcome to your Spanish lesson! Let’s practice. Can you translate 'I eat
breakfast at 7 am' into Spanish?"
Feedback Loop: The user attempts a translation, and the AI evaluates the response, providing
necessary corrections.

21
Adaptation: The AI offers tips based on the user's errors and introduces new phrases based on
the user's proficiency level.

Continuation: "Great job! How would you say, 'I will have dinner at 8 pm' in Spanish?"

Suggested input style: Free-form text encourages the organic learning and correction cycles that
are key in language learning.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best suited for scenarios that mimic human conversation, such as customer service or
educational tools, where the user interaction should feel natural and engaging.

It is ideal for use cases where the AI needs to adapt its language, tone, or content based on the
unfolding dialogue.

When to Use JSON:

It is useful when the interaction requires integrating with databases or other systems, such as in a
customer support scenario where transaction details or customer records need access.

Effective for structuring dialogue steps or storing user responses for analysis, especially when
detailed tracking of dialogue progress or user preferences is needed.

Conclusion:

The Interactive Prompts framework is a powerful tool for creating AI applications that require
sophisticated, dynamic interactions. AI can provide more personalized and contextually
appropriate responses by employing a conversational model that evolves through real-time user
feedback. Whether using Free Form Text for more natural dialogues or JSON for more structured
interactions, this framework enhances user engagement and satisfaction, making AI interactions
more productive and enjoyable.

22
Section 8: The TRACI (Task, Role, Audience, Create, Intent) Framework
The TRACI framework is a structured approach to designing AI prompts that ensures they are
tailored to specific contexts and objectives. This framework is ideal for creating highly targeted
interactions with AI systems, making it particularly valuable in content creation, customer
interaction, and educational applications.

Understanding the TRACI Components

TRACI stands for:

1. Task: Define what the AI is supposed to accomplish.

2. Role: Specify the role the AI assumes in this interaction.

3. Audience: Identify the intended audience for the AI’s output.

4. Create: Detail what precisely the AI needs to produce.

5. Intent: Clarify the purpose or goal behind the AI’s task.

Examples of the TRACI Framework in Use

Example 1: Marketing Content Creation

Task: Generate a blog post.

Role: AI acts as a content creator specializing in digital marketing.

Audience: The target audience is small business owners looking to increase their online
presence.
Create: A detailed, SEO-optimized blog post about cost-effective marketing strategies.

Intent: To educate the audience and drive engagement on digital marketing platforms.

Suggested input style: JSON allows for the structured input of parameters such as audience
demographics and content goals, facilitating precise and tailored content creation.

Example 2: Educational Tool for History Students

Task: Create an interactive quiz on 20th-century world history.

Role: The AI serves as an educational tool.

Audience: High school students studying world history.

Create: A set of multiple-choice questions with detailed explanations for each answer.

Intent: To test knowledge and reinforce learning through detailed feedback.

23
Suggested input style: Free-form Text allows the AI to generate educational content in a more
engaging and accessible format, fostering a better learning experience.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is suitable for scenarios where the output needs to be engaging and easily digestible, such as
narratives, educational content, or general customer interactions, where the tone and style should
feel natural and conversational.

It is ideal for tasks where flexibility in language and presentation is essential.

When to Use JSON:

It is perfect for applications requiring high customization and specificity, where various aspects
of the task, role, audience, creation, and intent must be clearly defined and programmatically
accessed.

It is helpful in environments where the AI’s output needs to integrate with other systems, such as
content management systems or customer relationship management platforms, where structured
data is paramount.

Conclusion:

The TRACI framework is an essential tool for developers and AI practitioners looking to craft
precise, goal-oriented AI prompts that are finely tuned to the needs of a specific audience. By
methodically defining the task, role, audience, creation, and intent, TRACI prompts to ensure
that AI interactions are adequate, relevant, and targeted. Whether employing Free Form Text for
a more fluid and adaptive interaction or using JSON for structured and detailed prompt design,
TRACI facilitates enhanced control over AI outputs, making it a cornerstone technique in AI
prompt engineering.

24
Section 9: The TRAACI (Task, Role, Analyze, Audience, Create, Intent)
Framework
The TRAACI framework builds upon the foundational TRACI model by incorporating an
additional component: Analyze. This addition enhances the framework's ability to handle
complex scenarios where preliminary analysis is crucial before generating a response. TRAACI
stands for Task, Role, Analyze, Audience, Create, and Intent, offering a comprehensive structure
instrumental in applications requiring detailed data processing or decision-making before action.

Understanding the TRAACI Components

TRAACI involves:

1. Task: Define the specific job or problem the AI must solve.

2. Role: Specify the function or identity the AI assumes in this scenario.

3. Analyze: Introduce a step where the AI must analyze or process information before creating an
output.
4. Audience: Identify the target audience for whom the output is intended.

5. Create: Explain what the AI needs to produce.

6. Intent: Clarify the goal or purpose of the task.

Examples of the TRAACI Framework in Use

Example 1: Financial Analysis Tool

Task: Evaluate potential investment opportunities.

Role: The AI acts as a financial analyst.

Analyze: Assess risk factors and return profiles based on current market data.
Audience: Financial advisors and their clients.

Create: A detailed report summarizing the analysis with recommendations.

Intent: To provide informed investment advice to enhance portfolio performance.

Suggested input style: JSON because it allows for detailed structuring of financial data and
parameters, enabling precise and systematic analysis.

Example 2: Healthcare Diagnostic Assistant

Task: Diagnose patient symptoms to suggest possible conditions.

Role: The AI serves as a diagnostic tool for healthcare professionals.

25
Analyze: Review symptoms, medical history, and relevant clinical guidelines.

Audience: Doctors and medical staff.

Create: A list of possible diagnoses along with the reasoning for each.

Intent: To aid healthcare professionals in making accurate diagnoses faster.

Suggested input style: JSON is ideal for organizing complex medical data and ensuring all
necessary information is systematically analyzed and presented.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best for applications that require a natural, conversational interface, such as customer service
bots or virtual assistants, where the interaction needs to mimic human dialogue.

It is helpful in less structured environments where the emphasis is on generating human-like text
or rigid data structuring is not critical.

When to Use JSON:

It is essential for tasks that demand detailed, structured input and output, such as in technical
fields where data must be processed or analyzed methodically.

When integration with other systems or databases is preferred, each aspect of the analysis and
creation process must be meticulously documented and accessible programmatically.

Conclusion:

The TRAACI framework provides a robust structure for designing AI prompts that require an
additional layer of analysis before generating outputs, making it ideal for complex and data-
intensive applications. TRAACI ensures that AI interactions are efficient and deeply aligned with
specific user needs and contexts by systematically defining the task, role, analysis, audience,
creation, and intent. Whether leveraging Free Form Text for more dynamic and flexible
interactions or employing JSON for structured and detailed engagements, TRAACI enhances the
AI’s capability to deliver precise and relevant solutions across various industries.

26
Section 10: The Scaffolded Prompts Framework
The Scaffolded Prompts framework is designed to guide the AI through a task by providing a
series of prompts that gradually build upon each other, leading to a complex response or action.
This approach is like educational scaffolding, where a teacher provides successive levels of
temporary support to help students achieve higher levels of comprehension and skill acquisition.

Understanding the Scaffolded Prompts Components

The framework typically involves:

1. Introduction: Setting the initial context or presenting the problem.

2. Development: Offering incremental information or questions that lead the AI towards a more
refined understanding.

3. Climax: Providing a prompt requiring the AI to synthesize the information in previous steps
and produce a comprehensive response.

4. Conclusion: A final prompt that might ask the AI to reflect, summarize, or apply the learned
information in a new context.

Examples of the Scaffolded Prompts Framework in Use

Example 1: Learning a New Concept

Introduction: "What is photosynthesis?"

Development: "How does photosynthesis contribute to the oxygen supply?"

Climax: "Can you explain the entire process of photosynthesis, from light absorption to oxygen
release?"

Conclusion: "How would the process of photosynthesis be different in a plant on Mars, if at

all?"
The suggested input style is free-form Text, which encourages a natural progression in the
explanation and is more suitable for educational contexts.

Example 2: Problem Solving in Customer Service

Introduction: "A customer reports their internet is down."

Development: "What steps would you take to troubleshoot this issue?"

Climax: "After following these steps, the router is malfunctioning. How do you proceed?"

Conclusion: "The customer is frustrated; how would you reassure them while solving the
problem?"

27
Suggested input style: JSON may be used to structure the steps and potential solutions,
especially if the AI’s responses require integration with a diagnostic tool or customer database.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best suited for scenarios where the interaction is more conversational or narrative-driven,
such as storytelling, teaching, or any context where a natural flow of information is essential.

It is ideal for applications where prompts are less about data retrieval and more about exploring
concepts or generating creative content.
When to Use JSON:

This is useful in technical scenarios where each scaffold step might require the AI to perform
specific actions or retrieve data, such as programming or complex problem-solving tasks.

This method is preferred when the AI's output at each step needs to be structured for further
processing or when the prompts are part of a more extensive automated system.

Conclusion:

The Scaffolded Prompts framework allows for a structured yet flexible approach to guide AI
through complex tasks and learning processes. By carefully crafting each scaffold stage,
developers and educators can ensure that the AI understands each part of the task and can
integrate all parts into a comprehensive response. Whether using Free Form Text to promote a
more fluid and natural interaction or JSON for structured and detailed engagements, this
framework supports a progressive buildup of knowledge and skills that can be adapted to various
applications.

28
Section 11: The SMART Framework
The SMART framework is a well-established tool originally used for setting goals and
objectives. It has been adapted for use in AI prompt engineering. SMART stands for Specific,
Measurable, Achievable, Relevant, and Timebound. Each element creates clear, actionable
prompts that guide the AI in producing targeted and effective results.

Understanding the SMART Components

1. Specific: The prompt should be clear and specific, leaving no ambiguity about what is
expected from the AI.

2. Measurable: There should be an explicit criterion for measuring the success of the AI's
response.

3. Achievable: The prompt should set realistic and attainable goals for the AI within its
capabilities.

4. Relevant: The task should be relevant to the AI's purpose and the broader objectives it's meant
to achieve.

5. Timebound: If applicable, there should be a timeframe within which the task should be
completed.

Examples of the SMART Framework in Use

Example 1: Content Moderation

Specific: "Identify and flag posts containing hate speech from the dataset."

Measurable: "Provide a count and percentage of flagged posts."

Achievable: "Use the trained hate speech detection model to evaluate the posts."

Relevant: "This task is part of our community's efforts to maintain a respectful online
environment."

Timebound: "Complete the task within 24 hours."

Suggested input style: JSON, as it allows for a structured approach where measurable
outcomes, such as the count and percentage, can be clearly defined and tracked.

Example: 2: Sales Report Generation

Specific: "Generate a weekly sales report for the Northeast region showing product-wise
breakdown."

Measurable: "The report should highlight total sales, bestselling products, and percentage
growth from the previous week."

29
Achievable: "Utilize the sales database to extract the necessary information."

Relevant: "This report aids the sales team in understanding regional performance."

Timebound: "Deliver the report every Monday morning."

Suggested input style: JSON organizes the report's parameters and automates the extraction of
sales data in a structured format.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

This is ideal for simpler tasks where prompts are straightforward and do not require extensive
structuring, such as general queries or when the AI is asked to generate narrative text.

Suitable for interactive scenarios where prompts are expected to flow naturally and can be
iteratively refined based on the AI's responses.

When to Use JSON:

This is crucial for tasks that involve specific criteria and quantifiable metrics, ensuring that each
component of the SMART framework is addressed accurately.

This is essential for scenarios requiring integration with systems that track and measure
outcomes or when the tasks involve multiple layers of data that the AI needs to process
systematically.

Conclusion:

The SMART framework brings a disciplined and results-oriented approach to AI prompt

engineering. By incorporating specific, measurable, achievable, relevant, and timebound
elements into prompts, users can significantly improve the efficacy and precision of AI-generated
outputs. Whether using the structured and detailed JSON format for complex tasks or opting for
the simplicity of Free Form Text for more straightforward inquiries, the SMART framework
ensures that AI tasks are goal-focused and practical, making it a valuable strategy for many AI
applications.

30
Section 12: The Prompt Chaining Framework
The Prompt Chaining framework is a method of prompt engineering that utilizes a sequence of
related prompts to guide AI models through a multistep process or a complex problem-solving
task. Each prompt in the chain builds on the last, allowing for progressive refinement and
expansion of the model's outputs.

Understanding the Prompt Chaining Components

Prompt Chaining involves:

1. Initial Prompt: Sets up the primary task or question for the AI to address.

2. Subsequent Prompts: Follow-up prompts that develop the initial response, asking the AI to
expand, clarify, or apply the information provided in its previous response.

3. Concluding Prompt: This is the final step in the chain, aiming to connect all previous
responses into a cohesive outcome.

Examples of the Prompt Chaining Framework in Use

Example 1: Recipe Development

Initial Prompt: "List the ingredients needed for a vegan chocolate cake."

Subsequent Prompts: "Now, describe the steps to prepare the batter with these ingredients,"
followed by, "Next, explain the process for baking and assembling the cake."

Concluding Prompt: "Finally, suggest a presentation style for serving the cake at a party."
Suggested input style: Free-form text can be more natural for this creative, sequential task,
miming how a human would approach recipe development.

Example: 2: Software Troubleshooting

Initial Prompt: "A user encounters an error when installing software X. What could be causing
this issue?"

Subsequent Prompts: "What steps can the user take to resolve this error?" followed by, "If the
error persists, what alternative solutions can we offer the user?"

Concluding Prompt: "Outline a comprehensive guide for troubleshooting this installation

error."

Suggested input style: JSON formatting can be beneficial here as it allows for structured inputs
at each stage, which can be critical for technical troubleshooting guides.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

31
Ideal for processes that benefit from a narrative style, such as creative writing, storytelling, or
any sequence where the output should flow like human discourse.

It is useful when the chain does not require handling complex data or when the flow from one
prompt to the next should feel seamless and conversational.

When to Use JSON:

It is best for technical sequences or problem-solving tasks where the AI must handle specific data
points or where the response at each step needs to be tracked and structured for further analysis.

This is useful when the output of each step will be used programmatically or requires clear
demarcation for processing by other systems or tools.
Conclusion:

Prompt Chaining is a dynamic and versatile framework that can lead to sophisticated AI-
generated content and solutions. By carefully designing a series of prompts that logically build
upon one another, users can navigate AI through intricate tasks that would otherwise be too
complex to tackle in a single step. The choice between Free Form Text and JSON formatting will
depend on the nature of the task and the desired flow of the interaction, with each format offering
unique advantages for different types of prompt chains.

32
Section 13: Contextual Prompting Framework
The Contextual Prompting framework enriches AI prompts with detailed background
information, allowing the AI to generate responses that are deeply informed by a specific
context. This framework is critical for tasks that require nuanced understanding or where the AI's
output needs to reflect a deep level of situational awareness.

Understanding the Contextual Prompting Components

Contextual Prompting typically involves:

1. Context Setting: Introducing the scenario or background information relevant to the task.

2. Task Specification: Clearly state what the AI needs to accomplish within the context.

3. Guidance Provision: Offering additional instructions or details to help shape the AI's output
according to the context.

Examples of the Contextual Prompting Framework in Use

Example 1: Historical Article Writing

Context Setting: "The year is 1920, and the United States has just ratified the 19th
Amendment."

Task Specification: "Write an article from a journalist's perspective during that time, discussing
the amendment's impact."

Guidance Provision: "Consider the social attitudes of the era and the suffrage movement's
journey."

Suggested input style: Free Form Text, as it includes rich, narrative details that bring the
historical context to life in the article.

Example 2: Customer Support for a Tech Product

Context Setting: "A customer has just purchased our latest smart home device but is struggling
with its setup."

Task Specification: "Provide a supportive response that guides them through the setup process."

Guidance Provision: "Include troubleshooting tips specific to common setup issues reported by
other users."

Suggested input style: JSON, particularly if the AI's response needs to pull in dynamic data
points from a knowledge base or user manual, ensuring the information is accurate and
contextually relevant.

Choosing Between Free Form Text and JSON Formatting

33
When to Use Free Form Text:

It is ideal for applications where the prompt requires storytelling or a richly detailed narrative,
such as creative writing, roleplaying scenarios, or scenarios that require the AI to 'put itself in
someone else's shoes.'

It is suitable when the context can be captured in a paragraph or two of descriptive text that sets
the scene for the AI's task.

When to Use JSON:

It is necessary for tasks where context is defined by specific, structured data points, such as
generating reports that reflect current market data or user behavior analytics.
It is useful when the context involves variables that might change over time or when different
scenarios must be programmatically triggered based on certain conditions.

Conclusion:

Contextual Prompting is a nuanced and robust framework that significantly enhances the quality
of AI outputs by infusing them with relevant situational details. Whether utilizing the narrative-
friendly Free Form Text format or the structured and datacentric JSON format, this framework
ensures that AI systems can generate technically accurate, contextually rich, and engaging
outputs. Users can achieve higher relevance and precision in AI-generated content and solutions
by effectively incorporating context into AI prompts.

34
Section 14: The Contrastive Prompts Framework
The Contrastive Prompts framework involves crafting pairs of prompts delineating what to do
and what not to do. The AI is guided toward the desired output by contrasting correct and
incorrect responses or approaches. This method is beneficial for training AI models to understand
nuanced differences and avoid specific patterns or behaviors.

Understanding the Contrastive Prompts Components

The framework typically involves:

1. Positive Prompt: Directs the AI towards the desired action or response.

2. Negative Prompt: Shows an example of an undesired action or response.

3. Comparison: The AI is expected to analyze the contrast between the positive and negative
prompts and generate an output that aligns with the positive prompt.

Examples of the Contrastive Prompts Framework in Use

Example 1: Sentiment Analysis

Positive Prompt: "Identify the positive sentiment in the customer’s feedback."

Negative Prompt: "Avoid confusing neutral or negative sentiments as positive."

Comparison: "Given a statement, differentiate between positive, neutral, and negative

sentiments, ensuring accurate categorization."

Suggested input style: JSON is well suited here, as structured data can help clearly define what
constitutes positive, neutral, and negative sentiments for the AI.

Example 2: Environmentally Conscious Blogging

Positive Prompt: "Write a blog post highlighting the benefits of renewable energy sources."
Negative Prompt: "Do not include arguments that support nonrenewable energy sources."

Comparison: "Create content that persuasively discusses why renewable energy is preferable,
contrasting it with the drawbacks of nonrenewable sources."

Suggested input style: Free-form Text is appropriate for this creative writing task. It allows for a
fluid, adaptable narrative based on the contrast provided.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best for creative tasks or those requiring natural language generation, where the AI must
navigate between nuanced expressions or concepts that are not strictly data-driven.

35
It is ideal for applications where contrasting ideas can be presented in a narrative or
conversational format, enhancing the learning or decision-making process.

When to Use JSON:

It is crucial for tasks with specific criteria that can be programmatically evaluated, such as
filtering content, categorizing data, or enforcing compliance with rules or guidelines.

This is useful when prompts involve multiple conditional scenarios or when it’s important to
delineate apparent, structured guidance for the AI to follow.

Conclusion:
The Contrastive Prompts framework is a strategic method to teach AI models the distinction
between what is desired and what is to be avoided. By presenting contrasting scenarios, the AI
learns to discern and replicate the correct behaviors or responses, enhancing its decision-making
and content-generation capabilities. Whether using the storytelling ease of Free Form Text or the
clear, conditional structuring of JSON, Contrastive Prompts effectively refine AI performance
across a spectrum of tasks.

36
Section 15: The Meta Prompt Framework
The Meta Prompt framework is a reflective approach to prompt engineering that involves
creating prompts about crafting prompts itself. This recursive framework is designed to fine-tune
the AI's ability to generate effective prompts for generating specific responses or actions. It's an
advanced method that helps improve the AI's own prompt generation capabilities.

Understanding the Meta Prompt Components

The framework generally involves:

1. Prompt Creation: Task the AI with generating a prompt for a specified task.

2. Evaluation: Analyze the quality and effectiveness of the AI-generated prompt.

3. Revision: Provide feedback on improving the AI's prompt.

4. Iteration: Use the revised prompt to generate a new response, repeating the cycle as
necessary.

Examples of the Meta Prompt Framework in Use

Example 1: Educational Quiz Generation

Prompt Creation: "Generate a prompt that would lead an AI to create challenging yet fair
multiple-choice questions for a high school biology quiz."

Evaluation: Assess the difficulty and relevance of the generated questions.

Revision: Offer feedback on aligning the questions more closely with the curriculum.
Iteration: Adjust the prompt based on feedback and generate new questions.

Suggested input style: Free-form Text may be more suitable for this creative and iterative
process, miming the natural way educators approach quiz design.
Example 2: Marketing Campaign Strategy

Prompt Creation: "Create a prompt instructing an AI to devise a social media strategy for a new
product launch."

Evaluation: Determine the strategy's potential reach and engagement levels.

Revision: Provide insights into current social media trends to refine the strategy.

Iteration: Modify the initial prompt to incorporate these insights and reevaluate the new
strategy.

Suggested input style: JSON formatting could be advantageous in structuring each element of
the marketing strategy and incorporating data-driven insights into the revision process.

37
Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is ideal for recursive tasks that benefit from open-ended, natural language input, such as
creating prompts for generating narratives, dialogues, or content where tone and style are
essential.

It is helpful in scenarios where iterative refinement and feedback are part of a conversational
loop between the user and the AI.

When to Use JSON:

This is crucial for scenarios where the meta prompt requires detailed specifications best
structured in a clear, machine-readable format, such as generating code, structured data, or
prompts that interact with specific APIs.

JSON's structured nature makes it easier to parse and analyze changes, effectively tracking the
evolution of prompts and responses across iterations.

Conclusion:

The Meta Prompt framework elevates the art of prompt engineering by turning the AI into a
collaborator in a prompt design, allowing for the development of more effective prompts through
a reflective and iterative process. Whether leveraging the fluidity of Free Form Text or the
structured precision of JSON, this framework provides a means to continuously refine the AI’s
prompting abilities, ensuring that the prompts lead to more accurate and purposeful AI-generated
outputs. By employing meta prompts, developers and AI practitioners can create a loop of
continuous improvement in AI systems' responsiveness and creativity.

38
Section 16: The Dynamic Prompting Framework
The Dynamic Prompting framework is a real-time interactive approach where AI prompts are
continuously adjusted based on the AI's responses or external inputs. This method allows for a
responsive dialogue where each prompt is informed by the latest information, making it
particularly suited for applications that require adaptability and immediate feedback.

Understanding the Dynamic Prompting Components

The framework typically involves:

1. Initial Prompt: Establishes the first point of interaction, setting up the task or question for the
AI.

2. Real-time Adaptation: This involves adjusting subsequent prompts in response to the AI's
outputs or changes in the environment or data.

3. Feedback Integration: This technique incorporates immediate feedback into the following
prompt to refine the AI's understanding or action.

4. Progressive Complexity: This method increases the complexity of tasks or the depth of
interaction as the AI completes each stage.

Examples of the Dynamic Prompting Framework in Use

Example 1: Interactive Storytelling

Initial Prompt: "Start a story set in a medieval fantasy world where the hero embarks on a
quest."

Realtime Adaptation: Based on the story's direction, prompts adapt to introduce new characters,
challenges, or plot twists.

Feedback Integration: Audience reactions or choices are used to guide the next segment of the
story.
Progressive Complexity: The narrative becomes more intricate as the AI develops the plot
further.

Suggested input style: Free Form Text facilitates a creative narrative's natural flow and
evolution.

Example 2: Customer Service Interaction

Initial Prompt: "Assist a customer having trouble logging into their account." Realtime
Adaptation: The following prompt may ask for alternative solutions if the AI's solution does not
resolve the issue.

Feedback Integration: Customer responses guide the AI's subsequent troubleshooting steps.

39
Progressive Complexity: As the issue gets resolved, prompts move towards ensuring customer
satisfaction and possibly upselling.

Suggested input style: JSON could be used if the interaction requires accessing and updating
customer records or real-time transaction logs.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

Ideal for scenarios that demand a conversational style, such as chatbots or interactive fiction,
where the dialogue needs to adapt organically based on the user's inputs.
It is suitable when the prompts and responses are less about data manipulation and more about
providing information, guidance, or entertainment.

When to Use JSON:

This is essential when dynamic prompts depend on constantly updated data, such as stock prices,
weather reports, or user behavior analytics.

It is beneficial for applications that require tracking multiple variables throughout an interaction
or where responses must trigger specific actions or updates to a system.

Conclusion:

The Dynamic Prompting framework represents a significant advancement in interactive AI,

providing a real-time model for engagements that evolve based on a continuous loop of input and
feedback. By utilizing Free Form Text for fluid interactions or JSON for data-driven scenarios,
Dynamic Prompting ensures that AI systems can respond with relevance and adaptability,
offering personalized and context-aware experiences. Whether the task is narrative or requires
strict adherence to rapidly changing data, this framework equips users with the tools to maintain
meaningful and effective AI-driven dialogues.

40
Section 17: The Multimodal Prompts Framework
The Multimodal Prompts framework is an advanced approach in AI prompt engineering that
integrates multiple types of inputs, such as text, images, audio, and video, to generate more
prosperous and comprehensive AI responses. This framework is critical when combining
different data forms to provide a complete understanding and produce more nuanced and
contextually relevant outputs.

Understanding the Multimodal Prompts Components

Multimodal Prompts usually involve:

1. Data Integration: Combining various data types, such as text descriptions with images or
audio clips.

2. Prompt Synthesis: Creating prompts that leverage the integrated data to instruct the AI on the
desired task.

3. Response Generation: The AI utilizes the combined data to generate a response that reflects
the complexity of the multimodal input.

4. Output Formatting: Ensuring the AI response is appropriately formatted to suit the

multimodal nature of the input.

Examples of the Multimodal Prompts Framework in Use

Example 1: Educational Content Creation

Data Integration: Combine a textbook excerpt with diagrams and graphs related to the content.

Prompt Synthesis: "Create an interactive lesson that explains the concepts in this text and
illustrates them using the attached visuals."

Response Generation: The AI develops a lesson plan that includes the explanatory text and
suggestions on using the visuals.
Output Formatting: The lesson is formatted into a slide presentation with annotations.

Suggested input style: JSON is preferable here as it effectively structures and references the
different types of content and instructions for lesson creation.

Example: 2: Social Media Campaign

Data Integration: Pair a product description with images of the product and customer reviews.

Prompt Synthesis: "Develop a social media post highlighting the product's features,
incorporating customer testimonials, and using the provided images."

41
Response Generation: The AI crafts a social media post with text, images, and curated customer
quotes.

Output Formatting: The post is formatted to comply with the specific social media platform's
requirements.

Suggested input style: While JSON could structure the different elements, Free Form Text may
also be suitable if the platform accepts composite posts in a straightforward, text-based format.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is ideal for scenarios where the AI's response does not require extensive data structuring, such
as generating simple social media posts or brief content descriptions that include multimodal
elements.

This is useful when the AI platform supports direct input of combined text and media, allowing
for a seamless creation process.

When to Use JSON:

This is crucial for scenarios that require detailed organization of multiple data types, mainly
when each data type must be processed differently, such as in complex content management
systems.

It is necessary when the prompt must specify particular relationships between the different
modalities or when certain actions must be taken with specific data types.

Conclusion:

The Multimodal Prompts framework opens a world of possibilities for AI interactions by

allowing for the integration of various data types into a single, coherent prompt. This method is
incredibly potent in today's digital environment, where content often spans several formats and
mediums. Whether employing the detailed structuring capabilities of JSON or the simplicity of
Free Form Text, this framework empowers AI systems to provide responses that are as rich and
varied as the inputs they receive, ensuring outputs that are profoundly informed and contextually
attuned.

42
Section 18: The Ethical Prompting Framework
The Ethical Prompting framework is an approach in AI prompt engineering that emphasizes
creating socially responsible and culturally sensitive prompts. It aims to produce AI interactions
that respect ethical standards, avoid biases, and promote fairness and inclusivity.

Understanding the Ethical Prompting Components

Ethical Prompting involves:

1. Values Identification: Recognize the core values and ethical principles relevant to the task.

2. Bias Mitigation: Craft prompts that minimize biases, stereotypes, or harmful content.

3. Cultural Sensitivity: Include considerations for cultural differences and diversity.

4. Fairness Assurance: Ensure the prompts do not inadvertently favor one group or outcome
over another.

5. Impact Assessment: Evaluate the potential social impact of the AI's responses.

Examples of the Ethical Prompting Framework in Use

Example 1: AI Recruitment Tool

Values Identification: Promote equal employment opportunities for all candidates.

Bias Mitigation: "Assess candidates based on skills and experience without regard to gender,
race, or age."

Cultural Sensitivity: Recognize and respect the diverse backgrounds of applicants.

Fairness Assurance: Use the same criteria for evaluating all candidates.

Impact Assessment: Review the hiring results for patterns that could indicate bias.

Suggested input style: JSON is beneficial here as it allows for the structured input of
assessment criteria, ensuring a fair and unbiased evaluation process.

Example 2: Health Information Bot

Values Identification: Provide accurate health information while respecting user privacy.

Bias Mitigation: Avoid making assumptions about users based on their inquiries.

Cultural Sensitivity: "Address health concerns in a manner that respects various cultural
practices and beliefs."

Fairness Assurance: Ensure that the advice given is consistent and unbiased.

43
Impact Assessment: Monitor user feedback to assess whether the information appropriately
meets diverse user needs.

Suggested input style: Free-form text, as it allows for the generation of responses that are
nuanced and considerate of diverse user contexts.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is suitable for scenarios where the AI's responses must be nuanced and adaptable to ethical
considerations, such as public-facing content or user interactions where tone and sensitivity are
crucial.
It is useful when the prompts must be open-ended to account for diverse responses and ethical
considerations.

When to Use JSON:

This is essential for scenarios where specific ethical guidelines or criteria must be
programmatically defined and consistently applied across various tasks or interactions, such as
content filtering systems or compliance monitoring tools.

This is beneficial when responses require integration with databases or systems that contain
ethical guidelines or rules that must be followed strictly.

Conclusion:

The Ethical Prompting framework is critical in ensuring AI systems operate within the bounds of
ethical standards and social responsibility. It is designed to cultivate AI interactions that are
intelligent, conscientious, and considerate of the broad spectrum of human values and cultural
contexts. Whether through Free-Form Text's flexible and accommodating nature or the structured
and rule-based approach of JSON formatting, this framework is fundamental to aligning AI
technology with society's highest ethical aspirations.

44
Section 19: The Hierarchical Prompting Framework
The Hierarchical Prompting framework is a structured method for developing AI prompts that
mirrors the natural human cognitive process of moving from general to specific information.
This framework helps the AI understand and prioritize information, leading to responses that
respect the logical hierarchy of concepts or actions.

Understanding the Hierarchical Prompting Components

Hierarchical Prompting usually involves:

1. Broad Overview: Begin with a prompt establishing a wide context or a high-level overview
of the topic.

2. Drilling Down: Follow with prompts that gradually narrow the focus, asking for more
detailed information or actions related to the topic.

3. Specific Details: Culminate with prompts that seek precise information or require detailed
action, deeply nested within the context provided by earlier prompts.

Examples of the Hierarchical Prompting Framework in Use

Example 1: Business Strategy Formulation

Broad Overview: "Describe the current trends in the e-commerce industry."

Drilling Down: "Analyze how these trends affect small businesses specifically."

Specific Details: "Develop a targeted marketing strategy for a small business specializing in
handmade crafts that leverages these trends."

Suggested input style: JSON could be beneficial here, as it allows for layering the information
in a structured format that can help the AI organize its response hierarchically.

Example 2: Medical Diagnosis

Broad Overview: "What are the common causes of abdominal pain?"

Drilling Down: "Focus on the causes related to digestive issues."

Specific Details: "Provide a diagnostic approach for a patient with abdominal pain likely due to
a digestive issue, detailing the tests and examinations required."

Suggested input style: Free-form text can be appropriate for medical diagnostics, as it allows
for a more fluid development of reasoning akin to a doctor's thought process.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

45
This is ideal for scenarios where the AI’s response benefits from a narrative style, such as
educational content delivery or storytelling, where information is presented in a layered,
approachable manner.

Applicable when the hierarchy relates more to conceptual understanding than structured data
processing.

When to Use JSON:

It is crucial for tasks that involve complex, multitiered data or when each level of the hierarchy
might trigger different computational pathways or outcomes.
This is beneficial when the AI’s outputs must be fed into systems requiring delineated
information layers for further processing.

Conclusion:

The Hierarchical Prompting framework is pivotal in directing AI towards responses that clearly
understand the depth and structure inherent in complex subjects or tasks. By effectively
employing either the Free-Form Text format for natural progression or the structured
organization of JSON for data-intensive tasks, this framework allows users to elicit AI-generated
content that is logically ordered and contextually rich. Whether in strategic planning, educational
pedagogy, or technical analysis, Hierarchical Prompting ensures that AI systems can navigate
through layers of complexity to provide accurate and relevant outcomes.

46
Section 20: The Guided Imagery Prompts Framework
The Guided Imagery Prompts framework is designed to generate rich visual content or
descriptions by guiding the AI with detailed imagery-related instructions. This framework is
essential in fields like creative writing, art, and design, where vivid descriptions or visual
elements are crucial.

Understanding the Guided Imagery Prompts Components

The framework involves:

1. Imagery Detailing: Crafting prompts that provide specific and rich descriptions of the desired
image or scene.

2. Purpose Orientation: Defining the purpose or intent behind creating the image, influencing
the style and elements of the imagery.

3. Sensory Incorporation: Including sensory details that describe the visual aspects and the
sounds, textures, and emotions associated with the imagery.

4. Evocative Language: Language that evokes strong images or emotions facilitates the creation
of compelling and immersive visuals.

Examples of the Guided Imagery Prompts Framework in Use

Example 1: Fantasy Novel Illustration

Imagery Detailing: "Create: an illustration of a lush, ancient forest shrouded in mist, with
towering trees and a narrow, winding path."

Purpose Orientation: "The image should invoke a sense of mystery and enchantment suitable
for the cover of a fantasy novel."

Sensory Incorporation: "Include the details of vibrant greenery, the texture of bark and leaves,
and the soft whisper of the forest."
Evocative Language: "The forest is alive with the magic of old; every leaf and stone is steeped
in stories untold."

Suggested input style: Free-form text is suitable for such creative tasks, as it allows for the fluid
expression of rich and detailed descriptions.

Example 2: Marketing Campaign for a Travel Agency

Imagery Detailing: "Design a visual for a beach vacation spot that showcases crystal clear
water, white sandy beaches, and happy vacationers."
Purpose Orientation: "The visual should emphasize relaxation and adventure, enticing viewers
to book a summer holiday package."

47
Sensory Incorporation: "The sounds of waves, the warmth of the sun on the skin, and the
carefree laughter of travelers should be palpable."

Evocative Language: "Let the viewers feel the call of the sea as a sanctuary of blue, inviting
them to leave footprints in the sand."

Suggested input style: JSON may be used if the visual elements need to be tagged or described
in a structured way for digital content creation tools.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

Ideal for scenarios requiring creativity and the generation of descriptive content where the AI’s
response needs to paint a picture through words or ideas.

It is suitable for tasks where the emphasis is on the story or mood conveyed by the imagery
rather than the technical aspects of image composition.

When to Use JSON:

Essential for processes involving digital content creation where specific visual elements need to
be identified, tagged, and manipulated programmatically.

This is useful when the imagery needs to be cataloged with metadata or used in systems that will
process the visual information further, such as a content management system.

Conclusion:

The Guided Imagery Prompts framework is a dynamic tool that leverages the power of
descriptive language to Create vivid mental images and guide AI in visual content generation.
This framework taps into the AI's ability to interpret and synthesize detailed descriptions into
rich, evocative content. Whether through the artistic freedom of Free Form Text or the organized
precision of JSON, Guided Imagery Prompts can transform the way AI contributes to creative
projects, producing outcomes that resonate with audiences and fulfill specific creative objectives.

48
Section 21: The Recursive Prompts Framework
The Recursive Prompts framework involves prompts that self-reference or build upon their
previous iterations, effectively creating a loop of self-improvement in the AI's responses. This
method is akin to recursive function calls in programming, where the output of one prompt cycle
is fed back into the system as an input for the next, continually refining the quality and accuracy
of the AI's outputs.

Understanding the Recursive Prompts Components

Recursive Prompting typically includes:

1. Initial Prompt: Starts the recursive loop with a base prompt that sets the context or task.

2. Output Analysis: Evaluate the AI’s initial response for quality, relevance, and completeness.

3. Prompt Refinement: Adjust the original prompt based on the analysis to target specific areas
for improvement.

4. Feedback Loop: The refined prompt generates a new response, which is then analyzed and
refined continuously.

Examples of the Recursive Prompts Framework in Use

Example 1: Recipe Optimization

Initial Prompt: "Provide a recipe for a chocolate cake."

Output Analysis: Assess whether the recipe is easy to follow and produces a desirable cake.
Prompt Refinement: "Modify the recipe to make it suitable for baking at high altitudes."

Feedback Loop: Use the adjusted recipe to bake a cake and refine it based on the outcome.

Suggested input style: JSON could be valid here, especially if the recipe adjustments require
precise measurements and conditions that must be systematically tracked.
Example: 2: Interactive Learning

Initial Prompt: "Explain Newton's first law of motion."

Review the AI’s explanation for clarity and comprehensiveness.

Prompt Refinement: "Expand the explanation to include everyday examples that illustrate the
law."

Feedback Loop: Analyze the expanded explanation and refine it to focus on areas that need
further clarification.

49
Suggested input style: Free-form text is appropriate here to allow the AI to build upon its
explanation naturally and educationally gradually.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is best for applications where the progression of the prompts should mimic natural, iterative
learning or creative processes, such as writing or educational content development.

It is useful when the responses require a degree of subjective assessment and refinement that can
only easily be translated into structured data.
When to Use JSON:

It is essential when the iterative process involves specific parameters or variables that need to be
altered and tracked over multiple iterations, such as optimizing code or refining a data analysis
algorithm.

It is beneficial in scenarios where the feedback loop results in discrete, quantifiable changes to
the input prompts, requiring clear documentation of each adjustment.

Conclusion:

The Recursive Prompts framework offers a powerful method for achieving refined and targeted
AI outputs through continuous self-referential improvement. This framework is particularly
beneficial in scenarios that require iterative development and the gradual enhancement of
responses. Whether through the narrative flexibility of Free Form Text or the structured clarity of
JSON, recursive prompting allows for the evolution of AI-generated content toward greater
sophistication and accuracy with each loop.

50
Section 22: The Adaptive Learning Prompts Framework
The Adaptive Learning Prompts framework is a responsive method of prompt engineering that
tailors the difficulty and content of prompts based on the AI's performance or the user’s progress.
This framework is particularly effective in educational technologies and personalized user
experiences where growth and development are essential.

Understanding the Adaptive Learning Prompts Components

The Adaptive Learning Prompts framework typically includes:

1. Assessment: Begin with a prompt that assesses the current level of understanding or
performance.

2. Adjustment: Based on the response, adjust the following prompt to increase or decrease
complexity as appropriate.

3. Progress Monitoring: Continuously monitor responses to determine the effectiveness of the

prompts and the progress being made.

4. Feedback Integration: Incorporate direct feedback into subsequent prompts to personalize

the learning path or user experience.

Examples of the Adaptive Learning Prompts Framework in Use

Example 1: Language Learning Application

Assessment: "Translate the following simple sentences from English to Spanish."

Adjustment: If the translations are correct, the prompt may involve more complex sentences; if
not, the following prompt will simplify.

Progress Monitoring: Keep track of the user's success rate over time.

Feedback Integration: "Since you've mastered present-tense verbs, let's practice past-tense
verbs now."

Suggested input style: JSON is advantageous for systematically structuring the levels of
complexity and tracking user progress.

Example 2: Interactive Fitness Coach

Assessment: "How many pushups were you able to complete in the last session?"

Adjustment: Depending on the answer, the next workout may either increase the count or focus
on improving form.

Progress Monitoring: Record the number of pushups and user feedback after each session.

51
Feedback Integration: "You’ve improved your count; let’s introduce a new exercise to develop
your strength further."

Suggested input style: Free Form Text may be more natural here, allowing for a conversational
tone to motivate and guide the user through their fitness journey.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

This is ideal for user-facing interactions where the AI's prompts should read as natural and
humanlike, such as tutoring or personal coaching, where encouragement and guidance play a
significant role.
It is useful when flexibility is needed to adapt to the user’s responses in a way that feels organic
and personalized.

When to Use JSON:

It is necessary in scenarios where precise tracking of user responses and systematic adjustments
to prompt complexity are required, such as in adaptive testing platforms or skill development
tools.

It is beneficial when prompts must be dynamically generated based on quantifiable user data or
performance metrics.

Conclusion:

The Adaptive Learning Prompts framework ensures that AI systems provide a responsive and
personalized experience, continually adjusting to meet the user's evolving needs. Whether
through the seamless engagement of Free Form Text or the detailed customization possible with
JSON, adaptive prompting allows AI to support continuous improvement and learning in a user-
centered way, making it a valuable approach in a variety of adaptive applications.

52
Section 23: The CrossModal Prompts Framework
The CrossModal Prompts framework is an integrative approach designed for generating AI
outputs that cross the boundaries of different modalities, such as transitioning from text-based
inputs to visual or auditory outputs. This framework is essential in creating a seamless user
experience across various forms of media.

Understanding the CrossModal Prompts Components

The CrossModal Prompts framework often involves:

1. Input Modality: Define the type of input that will be provided to the AI, such as text, images,
audio clips, etc.

2. Target Modality: Specify the desired output modality that the AI is expected to generate,
which could differ from the input modality.

3. Transformation Guidance: Provide instructions on interpreting and transforming the input

into the target modality.

4. Contextual Anchoring: Offer context that anchors the transformation, ensuring that the AI's
output remains relevant and accurate to the initial input.

Examples of the CrossModal Prompts Framework in Use

Example 1: From Text Description to Image Generation

Input Modality: "A description of a sunset over the ocean."

Target Modality: Generate an image that visually represents the described scene.

Transformation Guidance: "Capture the warm hues of the sunset and the reflection on the
water."

Contextual Anchoring: Include elements such as the horizon, the sun dipping into the ocean,
and the gentle waves.

Suggested input style: JSON can be particularly effective for detailing each visual component
that should be included in the image.

Example 2: From Music to Emotive Text

Input Modality: A snippet of a classical piano piece.

Target Modality: Write a paragraph describing the emotions the music conveys.

Transformation Guidance: "Express the crescendos as moments of revelation and the softer
passages as introspective."

Contextual Anchoring: Relate the emotions to universal experiences like joy, sorrow, or hope.

53
Suggested input style: Form Text suits the task well, allowing for a fluid and expressive
translation of music into emotive text.

Choosing Between Free Form Text and JSON Formatting

When to Use Free Form Text:

It is suitable for scenarios where the transformation from one modality to another requires a
degree of interpretative freedom, such as literary or artistic tasks.

This is ideal when the input or output is narrative, requiring a seamless flow that a too rigid
structure might constrain.
When to Use JSON:

It is essential for tasks where the input consists of structured data that must be precisely mapped
to a different modality, such as converting scientific data into a graphical representation.

This is beneficial when the AI's output needs to be highly specific and might include metadata or
detailed instructions for creating content in the target modality.

Conclusion:

The CrossModal Prompts framework is a dynamic tool that facilitates the AI’s ability to operate
across different sensory and content modalities, broadening the scope of AI applications. By
guiding AI through interpreting inputs and generating contextually relevant outputs in entirely
new formats, this framework is instrumental in creating rich, multimodal user experiences.
Whether leveraging the descriptive power of Free Form Text or the precision of JSON,
CrossModal Prompts support the creation of AI outputs that are as varied and nuanced as the
inputs they derive from.

54
Closing Section: Embracing the Future with Advanced Prompt Engineering
As we conclude this journey through the intricate landscape of Advanced Prompt Engineering,
it's essential to reflect on the transformative power of carefully crafted prompts in unlocking the
full potential of AI. The frameworks explored in this guide—from the structured clarity of
GRADE to the creative depths of Guided Imagery Prompts—represent a confluence of precision
and creativity, necessity and innovation, forming the backbone of effective AI interactions.

The role of prompts extends beyond mere input; they are conversations that shape the AI’s
understanding and outputs. The thoughtful application of frameworks like Hierarchical
Prompting or the adaptive finesse of Recursive Prompts can elevate AI from a tool to a
collaborator. With the insights from the Ethical Prompting framework, we recognize our
responsibility to guide AI along a path that upholds our highest ethical standards, ensuring
inclusivity and respect for all users.
Each framework we've discussed serves a unique purpose and applies to various domains.
Whether in education, creative industries, customer service, or beyond, the thoughtful
employment of these frameworks will lead to advancements in AI that were once the realm of
science fiction. CrossModal Prompts push the envelope, blurring the lines between the digital
and sensory worlds, enabling AI to engage with us through all our senses. Meanwhile, the
Adaptive Learning Prompts framework promises a future where AI personalizes growth and
learning, adapting to our individual needs with every interaction.

While seemingly technical, the choice between Free-Form Text and JSON formatting
underscores AI's dual nature as both an artistic and a scientific endeavor. The free-form text
allows us to interact with AI in our natural language, while JSON structures our most complex
thoughts into understandable bytes. The harmony between these two modalities is a dance of
functionality and form, reflecting the diverse ways we communicate and think.

As we harness these frameworks, it’s imperative to look forward with a sense of stewardship.
The Recursive Prompts framework isn't just a methodology; it’s a metaphor for our iterative
journey with AI, where the lessons inform each step forward of the past. Our partnership with AI
is not just about what it can do for us today but also how it can grow with us into tomorrow. As
we teach AI, we learn from the data, outcomes, and stories we create together.

This guide is not merely a collection of instructions but an invitation to innovate, dream, and
cooperate with AI. The vast expanse of possibilities is matched only by our ambition to explore
them. The future of AI prompt engineering is as boundless as our willingness to imagine and as
detailed as the prompts we construct. So, let us step into this future with intention, creativity, and
an unwavering commitment to ethical and responsible AI development.

Let the frameworks detailed in this guide be your compass as you navigate the evolving AI
landscape, crafting prompts that elevate technology into an art form—a symphony of human

55
intellect and machine capability. Embrace these tools and, with them, build a world where AI not
only answers our questions but also enriches our understanding and enhances our humanity.

56
Appendices: JSON Examples for Prompting Frameworks
Below, you will find JSON examples illustrating how each of the 23 prompting frameworks
discussed might be structured. These examples provide a foundation for understanding how to
implement these frameworks in practical AI applications.

Note: These are illustrative examples and may need to be adjusted for specific AI models and
platforms.

Appendix A: JSON Examples

1. GRADE Framework JSON Example

```json
{
"goal": "Generate an informative blog post",
"request": "Write about the impact of AI on modern healthcare",
"action": "Include latest AI advancements, statistics, and
expert quotes",
"details": {
"wordCount": 1200,
"audience": "Healthcare professionals",
"keywords": ["AI", "healthcare", "technology"]
},
"example": "AI is revolutionizing healthcare by..."
}

57
2. RODES Framework JSON Example
```json
{
"role": "Market analyst",
"objective": "Provide a stock market trend analysis",
"details": {
"focus": ["Technology sector", "Q1"],
"dataSources": ["NASDAQ", "Bloomberg"]
},
"examples": ["Example of bullish trends", "Example of bearish
trends"],
"senseCheck": "Ensure data accuracy and relevance to current
economic events"
}

3. Chain of Thought (CoT) Framework JSON Example

```json
{
"task": "Solve a math problem",
"steps": [
"Identify the type of problem (e.g., algebra, geometry)",
"Break down the problem into solvable parts",
"Apply the appropriate mathematical rules or formulas"
],
"problem": "If a train leaves New York for Boston at 3 PM
traveling at 60 mph, and the distance is 200 miles, when will it
arrive in Boston?"
}

58
4. Zero-Shot and Few-Shot Learning Framework JSON Example
```json
{
"task": "Language translation",
"type": "Zero-shot",
"instruction": "Translate the phrase into French without prior
examples",
"phrase": "The quick brown fox jumps over the lazy dog."
}

5. ReAct (Reason and Act) Framework JSON Example

```json
{
"situation": "A customer received a defective product",
"reasoning": "Assess the situation based on the return policy",
"action": "Guide the customer through the return process"
}

59
6. Instruction Tuning Framework JSON Example
```json
{
"task": "Content moderation",
"criteria": [
"Filter out content that violates community guidelines",
"Check against a list of prohibited topics and language"
],
"feedback": "Increase sensitivity to aggressive language",
"adjustments": "Lower threshold for detecting hostility in
comments"
}

60
7. Interactive Prompts Framework JSON Example
{
"initialPrompt": "I see you're interested in planning a weekend
getaway! What kind of destination are you looking for - city,
countryside, or beach?",
"followUpPrompts": [
{
"userResponse": "beach",
"aiResponse": "Great choice! Do you prefer a quiet, secluded
spot or somewhere more lively and full of activities?"
},
{
"userResponse": "quiet, secluded spot",
"aiResponse": "Lovely! How far are you willing to travel?"
},
{
"userResponse": "within 200 miles",
"aiResponse": "Perfect, I have two recommendations for you:
Tranquil Tides Beach Resort and Sandy Shores Bungalows. Would
you like more details on these options?"
},
{
"userResponse": "more details",
"aiResponse": "Tranquil Tides Beach Resort offers luxury
accommodations with a private beach, spa services, and gourmet
dining. Sandy Shores Bungalows offers cozy, family-friendly
cabins with beachfront views. Which option interests you more?"
},
{
"userResponse": "Tranquil Tides Beach Resort",

61
"aiResponse": "Excellent choice! Would you like to check
availability for this weekend or get information on amenities?"
},
{
"userResponse": "check availability",
"aiResponse": "Checking now... Tranquil Tides Beach Resort has
availability for this weekend. Shall I go ahead and book a room
for you?"
}
],
"conclusion": {
"userResponse": "yes, book it",
"aiResponse": "Great, I've booked your stay at Tranquil Tides
Beach Resort. You will receive a confirmation email shortly. Is
there anything else I can help you with?"
}
}

62
8. TRACI Framework JSON Example
```json
{
"task": "Create a press release",
"role": "Public Relations Manager",
"audience": "Tech industry journalists",
"create": "A concise announcement about a new product launch",
"intent": "Generate media interest and coverage for the product
launch"
}

9. TRAACI Framework JSON Example

```json
{
"task": "Environmental impact report",
"role": "Environmental Analyst",
"analyze": "Review recent changes in legislation affecting
emissions",
"audience": "Government regulators",
"create": "A detailed report outlining compliance strategies",
"intent": "Ensure company practices align with new environmental
standards"
}

63
10. Scaffolded Prompts Framework JSON Example
```json
{
"intro": "Explain the concept of climate change",
"development": "Discuss the main causes contributing to climate
change",
"climax": "Describe specific effects of climate change on
coastal cities",
"conclusion": "Suggest mitigation strategies for reducing impact
in these areas"
}

11. SMART Framework JSON Example

```json
{
"specific": "Increase website traffic by 20% over the next
quarter",
"measurable": "Use analytics tools to track daily visitor
counts",
"achievable": "Implement SEO best practices and increase ad
spend",
"relevant": "Traffic growth supports the overall goal of
improving online sales",
"time_bound": "Complete within the next three months"
}

64
12. Prompt Chaining Framework JSON Example
```json
{
"initialPrompt": "Analyze the recent customer feedback for our
online retail store to identify common complaints.",
"chain": [
{
"step": 1,
"prompt": "Identify the top three recurring issues mentioned in
the customer feedback from the past month.",
"action": "Extract complaints from customer feedback data"
},
{
"step": 2,
"prompt": "Provide a detailed analysis of each identified issue,
including potential reasons for customer dissatisfaction.",
"action": "Analyze feedback to understand underlying causes"
},
{
"step": 3,
"prompt": "Suggest actionable solutions that can be implemented
to address each of the top three issues.",
"action": "Generate solutions based on analysis"
},
{
"step": 4,
"prompt": "Outline a plan to integrate these solutions into our
current operations, specifying any required changes to processes
or resources.",

65
"action": "Develop integration strategies for solutions"
},
{
"step": 5,
"prompt": "Propose methods for monitoring the effectiveness of
the implemented solutions in improving customer satisfaction.",
"action": "Create monitoring and evaluation strategies"
}
],
"conclusion": "Summarize the complete strategy to enhance
customer satisfaction, highlighting the expected impacts and
timeline for seeing improvements."
}

66
13. Prompt Chaining Framework JSON Example

{
"userProfile": {
"name": "John",
"age": 34,
"interests": ["history", "gastronomy", "outdoor activities"],
"previousDestinations": ["Italy", "Spain", "France"]
},
"travelDetails": {
"destination": "Greece",
"duration": "7 days",
"budget": "moderate",
"season": "spring"
},
"contextualPrompts": [
{
"context": "Considering John's interest in history, recommend
historical sites in Greece that are less crowded during the
spring season."
},
{
"context": "Given John's love for gastronomy, suggest a culinary
tour that includes traditional Greek dishes not commonly found
in tourist areas."
},
{
"context": "Based on his enjoyment of outdoor activities,
provide options for hiking or sailing that fit within his
moderate budget."

67
}
],
"finalTask": "Compile the suggestions into a day-by-day
itinerary that balances John's interests with ample downtime for
relaxation."
}

68
14. Contextual Prompt Framework JSON Example

{
"userProfile": {
"name": "John",
"age": 34,
"interests": ["history", "gastronomy", "outdoor activities"],
"previousDestinations": ["Italy", "Spain", "France"]
},
"travelDetails": {
"destination": "Greece",
"duration": "7 days",
"budget": "moderate",
"season": "spring"
},
"contextualPrompts": [
{
"context": "Considering John's interest in history, recommend
historical sites in Greece that are less crowded during the
spring season."
},
{
"context": "Given John's love for gastronomy, suggest a culinary
tour that includes traditional Greek dishes not commonly found
in tourist areas."
},
{
"context": "Based on his enjoyment of outdoor activities,
provide options for hiking or sailing that fit within his
moderate budget."

69
}
],
"finalTask": "Compile the suggestions into a day-by-day
itinerary that balances John's interests with ample downtime for
relaxation."
}

70
15. Meta Prompts Framework JSON Example
{
"initialPrompt": "Generate a question prompt for a high school
history lesson on Ancient Egypt that focuses on the construction
of the pyramids.",
"metaAnalysis": {
"evaluatePrompt": "Assess if the generated prompt effectively
engages students in critical thinking about architectural
techniques and the societal implications.",
"desiredCharacteristics": ["educational value", "engagement
level", "historical accuracy"]
},
"refinementSteps": [
{
"step": 1,
"feedback": "The prompt should encourage students to explore the
role of labor forces in ancient societies.",
"adjustment": "Incorporate questions about the social structure
and labor roles in the pyramid construction process."
},
{
"step": 2,
"feedback": "Include a comparative analysis aspect to enhance
critical thinking.",
"adjustment": "Modify the prompt to compare pyramid construction
techniques and purposes with those of other ancient
civilizations."
}
],

71
"finalTask": "Review the refined prompt and ensure it aligns
with curriculum standards for complexity and depth suitable for
high school students."
}

72
16. Dynamic Prompting Framework JSON Example
{
"initialPrompt": "Tell me about your current financial goals.",
"conversationFlow": [
{
"userResponse": "I want to save for retirement.",
"aiPrompt": "How many years away is your retirement, and do you
already have any savings or investments?"
},
{
"userResponse": {
"years": 20,
"currentInvestments": "Some"
},
"aiPrompt": "Considering you have some investments, would you like
to explore more aggressive investment options or prefer a
conservative approach?"
},
{
"userResponse": "I'm interested in aggressive investments.",
"aiPrompt": "Great choice for your timeline! Are you interested in
stocks, real estate, or perhaps cryptocurrencies?"
},
{
"userResponse": "Tell me more about stocks and cryptocurrencies.",
"aiPrompt": "Stocks offer potential growth through market
participation, whereas cryptocurrencies can offer higher returns
but with higher risk. Would you like a detailed comparison or
guidance on starting with either?"
},

73
{
"userResponse": "Yes, a detailed comparison please.",
"aiPrompt": "I'll provide an overview focusing on risk factors,
potential returns, and liquidity of both stocks and
cryptocurrencies."
}
],
"conclusion": {
"aiPrompt": "Based on our discussion, I recommend consulting with
a financial advisor to make the best choice for your situation.
Would you like help finding one?"
}
}

74
17. Multimodal Prompts Framework JSON Example
{

"prompt": {
"text": "I want to create a birthday card for my grandmother who
loves gardens.",
"visuals": {{

"theme": "garden",
"elements": ["flowers", "butterflies", "sunny sky"]
},
"instructions": "Design a card that includes these elements in a
vibrant, cheerful style."
},
"outputRequirements": {
"format": "image",
"size": "1080x720 pixels",
"colorScheme": "bright and colorful",
"textIncluded": {
"message": "Happy Birthday Grandma! Wishing you joy and
tranquility like your beautiful garden.",
"fontStyle": "handwritten",
"position": "bottom center"
}
},
"additionalContext": {
"audience": "elderly person who appreciates nature",
"occasion": "birthday",
"sentiment": "warm and loving"

75
}
}

76
18. Ethical Prompting Framework JSON Example
{
"task": "Generate a news article",
"topic": "Use of AI in healthcare",
"ethicalGuidelines": {
"accuracy": "Verify all AI claims with credible sources.",
"biasAvoidance": "Ensure the article does not favor any
political perspective or commercial interest.",
"privacy": "Do not include personal data or identifiable
information about patients without consent.",
"harmReduction": "Highlight both benefits and potential risks of
AI in healthcare."
},
"sources": [
{
"type": "academic journal",
"details": "Journal of AI Research, latest issue on healthcare
applications"
},
{
"type": "interview",
"details": "Talk with a leading ethicist in the field of AI"
}
],
"outputRequirements": {
"length": "800 words",
"tone": "informative and neutral",
"citations": {
"style": "APA",

77
"minimumNumber": 3
}
},
"additionalInstructions": "Include a section discussing ethical
considerations surrounding AI technologies in medicine, drawing
from the provided sources."
}

19. Hierarchical Prompting

{
"initialPrompt": "Provide an overview of the new government
policy on renewable energy subsidies.",
"hierarchy": [
{
"level": 1,
"detail": "Summarize the key components of the policy and its
primary objectives.",
"prompt": "What are the main elements of the policy and what
does it aim to achieve?"
},
{
"level": 2,
"detail": "Analyze the expected economic impact of the policy on
the renewable energy sector.",
"prompt": "How will the subsidies affect the growth and
sustainability of renewable energy companies?"
},
{
"level": 3,
"detail": "Evaluate the potential secondary effects on related
industries and overall national economy.",

78
"prompt": "What are the broader economic implications for
industries such as manufacturing and utilities?"
},
{
"level": 4,
"detail": "Propose measures to maximize the benefits and
mitigate any negative impacts of the policy.",
"prompt": "What strategies should be implemented to enhance the
positive effects and reduce any drawbacks?"
}
],
"conclusion": {
"prompt": "Consolidate all the information into a comprehensive
report detailing the policy's economic impact from a macro and
micro economic perspective."
}
}

79
20. Guided Imagery Prompts Framework JSON Example
```json
{
"description": "Create an image of a bustling medieval market",
"details": {
"visualElements": ["wooden stalls", "colorful fabrics", "diverse
crowd"],
"sounds": "Chatter of people, clinking of coins",
"smells": "Fresh bread, spices",
"mood": "Energetic and chaotic"
}
}

21. Recursive Prompts Framework JSON Example

```json
{
"initialPrompt": "Write a brief on renewable energy trends",
"evaluate": "Assess clarity and depth of information",
"refine": "Add recent statistics on solar energy adoption",
"iterate": "Incorporate expert opinions on future trends"
}

80
22. Adaptive Learning Prompts Framework JSON Example
```json
{
"initialPrompt": "Describe the process of photosynthesis",
"userResponse": "User provided a basic explanation",
"adaptivePrompt": {
"nextLevel": "Provide a detailed explanation including the
chemical reactions involved in photosynthesis",
"hints": "Consider the role of sunlight, water, and carbon
dioxide"
}
}

23. Cross-Modal Prompts Framework JSON Example

```json
{
"inputModality": "text",
"textInput": "A serene landscape description",
"targetModality": "image",
"transformationGuidance": "Convert the text description into a
high-resolution image",
"contextualAnchoring": {
"elements": ["mountains", "lakes", "forests"],
"mood": "peaceful"
}
}

81
Each JSON example is structured to facilitate the specific requirements of the respective
framework, providing a clear template for configuring prompts tailored to achieve defined
outcomes. These structured prompts help ensure that AI systems receive all necessary
information to perform tasks effectively, adhering to the guidelines established for each
framework.

Each example above is tailored to illustrate how the JSON format can structure the prompts for
different frameworks, ensuring that each aspect is clearly defined for the AI system. The JSON
structure provides a way to organize the prompts hierarchically, breaking down complex tasks
into manageable components that the AI can systematically process.

These examples encapsulate the essential structure and components needed to implement each
framework in JSON format. They ensure clarity of communication between users and AI
systems, providing a solid foundation upon which to build more complex interactions.

82
Appendix B: Resources
1. LLM’s, Prompt Libraries, and Resources

LLM's

Claude
ChatGPT
Gemini
Copilot

Prompt libraries

How to Build a Prompt Library for Business

What is a Prompt Library? And Why All Organizations Need One
Wolfram Prompt Repository for LLM AIs
GitHub - f/awesome-chatgpt-prompts: This repo includes ChatGPT prompt curation to use
ChatGPT better.
70 Best Prompts repository AI tools - 2024
The Prompt Index: Explore the Ultimate Free AI Resource
Creating an Effective Prompt Repository for Enhanced AI-Language Model Interactions | by
Krgoswami | Medium
PromptBase | Prompt Marketplace: Midjourney, ChatGPT, DALL·E, Stable Diffusion & more.
PromptHero - Search prompts for Stable Diffusion, ChatGPT & Midjourney
Wielded AI Productivity Tool ChatGPT Prompt Repository
Prompt Library - SinCode AI
Prompt Library — AI for Education
GoDaddy's AI prompt library — the best AI prompts for small businesses - GoDaddy Blog
The Ultimate AI Prompt Library for Content & SEO | Semrush
AI Prompt Library - Content at Scale
Copy.ai's AI Prompt Library: Create Better AI Content
Elementor AI's Prompt Library
Prompt Library – Free Prompts for Midjourney
Best AI Prompts for ChatGPT, Midjourney, Claude,…

AI certifications

Microsoft Certified: Azure AI Engineer Associate - Certifications | Microsoft Learn

AI-900 - Azure AI Fundamentals Study Guide - YouTube

Prompting techniques

Structured Prompt Notation Editor - Free to use, no login required.

LLM Prompting: The Basic Techniques | Determined AI

83
These additional resources can be valuable for anyone looking to expand their knowledge of
prompt engineering or find communities to share prompts and strategies.

As AI technology evolves, so will the methods and frameworks for prompt engineering. The
examples and resources provided here are starting points for a journey of exploration and
mastery in prompting AI systems. Whether you're looking to delve into the technical aspects of
JSON formatting or engage in the creative process of freeform prompt writing, the landscape is
rich with possibilities.

84