INTERACTIVE GUIDEPrompt Engineering 101

Foundations

Prompt Engineering & Why It Matters

A prompt is the input — instruction, command, or question — given to an AI model. The better the prompt, the better the output.

• Superior prompts yield higher-quality, more relevant responses
• Reduces time spent on revisions and refinements
• Transforms AI from a search engine into a specialist working for you

Example

Weak Prompt

"Write a business proposal."

Strong Prompt

"You are a B2B sales strategist. Write a concise, persuasive 2-page proposal for a $50,000 SaaS subscription service targeting mid-sized fintech companies. Use a professional yet compelling tone, and include ROI metrics and a closing call to action. Please ask up to 5 questions, one by one, to ensure you have the necessary details."

Key Takeaway: Defining who AI is, what it should do, and how it should format the response leads to far better results.

Structuring Prompts for Maximum Impact

Myzone's CREATE Framework

• Context Provide relevant background for better understanding
• Role Define AI's perspective for a tailored response
• Example Show the desired structure, tone, or format
• Ask Enable AI to request clarifications one at a time
• Task State precisely what output to generate
• Enhance Refine iteratively based on feedback

A high quality AI prompt includes all of the above.

Simple vs. Engineered Prompts

Simple Prompts

• Great for fast ideas and quick answers
• Ideal for lower-risk, low-stakes tasks
• Best when speed is the priority

Engineered Prompts

• Designed for complex tasks requiring structure
• Necessary when accuracy and consistency matter
• Appropriate for high-stakes decisions with impact

Context Engineering

What Is Context Engineering

Context engineering gives the AI the background it needs to understand your situation.

• Strong context reduces guessing and improves accuracy
• Clear context includes the purpose, audience, goals, and constraints
• Include only the details that matter — avoid information overload
• Frame context as if briefing a capable colleague who is new to your project. See Anthropic's guidance on briefing a colleague with minimal context.

Tip: For research tasks, part of your context can be: "First, go research this industry and list 5–10 reputable sources, then use those to inform your response."

How to Load Context Properly

Quality context anchors AI reasoning and significantly reduces errors.

• Include only the details that matter — keep it focused
• Share the purpose, audience, goals, and constraints clearly
• Avoid overwhelming the AI with irrelevant background information
• Frame context like briefing a human colleague who knows the field but not your situation
• Provide examples of the output format you want when possible

Prompt Engineering Process

Engineer Your Prompt Before You Use It

Most people write a prompt once and use it right away — which leads to inconsistent results. For high-value, complex, or repetitive work, engineer the prompt first.

• Open a separate chat where AI acts as your prompt engineer
• Have it apply the CREATE framework through an interview process
• Test the engineered prompt in a fresh chat or target tool
• Return to the engineering chat if revisions are needed
• Archive the final prompt for future reuse

Myzone's CREATE Framework — Detailed Definitions

• Context Supply relevant background information for better comprehension. Include purpose, audience, goals, and constraints.
• Role Establish AI's perspective for customized output. 'You are a seasoned CFO with 20 years of experience...'
• Example Demonstrate the desired structure, tone, or presentation. Show, don't just tell.
• Ask Enable AI to request one-by-one clarifications before proceeding — this catches gaps early.
• Task State precisely what output to generate. Be explicit about format, length, and deliverable.
• Enhance Refine iteratively based on feedback. Return to the engineering chat to improve the prompt, not the output.

CREATE Framework in Action — Digital Marketing Example

• C – Context "A tech startup is launching an AI-driven CRM tool and needs a structured content marketing plan to reach and engage its target audience effectively."
• R – Role "You are a seasoned digital marketing strategist with over 10 years of experience, specializing in tech startups."
• E – Example "The final output should include a detailed plan covering audience segmentation, recommended content types, and distribution strategies."
• A – Ask "Before proceeding, ask up to 5 clarifying questions, one at a time, to ensure you have the necessary details to create a precise and tailored strategy."
• T – Task "Based on the gathered insights, develop a comprehensive content marketing plan for the AI-driven CRM startup, ensuring it aligns with best practices and market trends."
• E – Enhance "This is a reminder to you (the prompter, not the AI) that most prompts are not nailed the first time. Return and iterate."

Final Prompt (Following CREATE Framework)

You are a seasoned digital marketing strategist with over 10 years of experience, specializing in tech startups. A new AI-driven CRM tool is launching, and the company needs a structured content marketing plan. Your task is to develop a plan that includes audience segmentation, content types, and distribution strategies. Before starting, ask up to 5 clarifying questions, one at a time, to ensure you have enough information. Once you've gathered enough information, provide a detailed strategy and be prepared to refine it based on feedback.

Why It Works: AI takes on the role of an expert and generates insights as if it were a specialist.

Try It: Assign AI different roles (e.g., "You are a CFO" or "You are an HR consultant") and see how the responses change.

Prompt Engineering Workflow

1. Draft a simple version of your prompt Start with the core ask — what do you actually want?
2. Open a separate engineering chat Ask the AI to act as your prompt engineer and apply CREATE through an interview.
3. Test in a fresh chat or target tool Use a fresh chat when you want to isolate the prompt from earlier context; also test in its intended workflow.
4. Review the output and identify gaps What's missing, wrong, or off-tone? Be specific.
5. Return to the engineering chat Share what you got and what you want changed. Ask it to enhance the prompt.
6. Repeat until results are reliable Prompt engineering is a loop, not a one-shot process.
7. Save the final prompt to your library Once it's reliable, archive it for future reuse and team sharing.

Prompt engineering is a loop that strengthens your prompt each time.

How to Test a Prompt

Once your prompt has been engineered, test it with a real use case before adding it to your library.

• Does the output match the goal and audience of the task?
• Does the structure follow what you asked for?
• What is missing, wrong, or off-tone?
• Would the output be usable without substantial revision?

How to Enhance Your Prompt

Enhancement is different from editing the output — you improve the prompt, not the result.

• Return to the separate engineering chat — not the task chat
• Paste in the output you got and describe what you want changed
• Tell the AI to stay in the role of prompt engineer
• Ask it to explain what modifications it's making and why
• Retest in a fresh session and either finalize or continue iterating

Advanced Techniques

Chain of Thought Prompting

Step-by-step prompting is one technique to try, not a requirement for every model or task. Google Research's 2022 GSM8K math report gives 58% accuracy for PaLM with chain-of-thought prompting and an external calculator, versus the prior 55% result for fine-tuned GPT-3 with a verifier.

• Useful for planning, analysis, problem-solving, and strategy work. The same 2022 Google Research report gives 74% GSM8K accuracy for a follow-up self-consistency method that votes across multiple reasoning paths.
• Ask for a concise explanation and checkable intermediate results when useful; follow the model guidance for reasoning tasks.
• Check the answer against evidence; an explanation alone does not establish accuracy.
• Historical benchmark results do not guarantee gains with current models.

Try it: Add "Think step by step" to your next complex prompt and compare the output quality.

What Are Hallucinations?

A hallucination is when AI gives a confident answer that is simply wrong or made up.

• Invented facts, fake citations, or completely fabricated information
• Outdated data presented as current and accurate
• Results from knowledge gaps or overconfident pattern-matching
• Modern models reduce but cannot eliminate hallucination risk — always verify

How to Minimize AI Misinformation

You can dramatically reduce hallucination risk with the right prompting strategies.

• Use fact-based modes for research — Perplexity AI, Deep Research in ChatGPT Pro
• Phrase prompts to demand accuracy: "Cite reputable sources and list the publication and date for each"
• Run a source research step before the main task: "First find 5–10 reputable sources, then use those to inform your response"
• Double-check AI outputs against trusted primary documents or official sites. Microsoft's prompt guidance also says model responses still need validation.

Tip: Ask the AI: "What are best practices for reducing hallucinations on this topic?" and follow its suggestions as part of your workflow.

Human in the Loop for High-Stakes Work

For high-stakes decisions, always have an expert review AI outputs. AI is an efficient research assistant — not a final decision authority.

• Legal contracts and compliance matters require qualified expert review
• Financial commitments should be verified by a financial professional
• Medical or HR guidance must be reviewed by certified practitioners
• Higher stakes = greater human review importance — no exceptions

Choosing the Right Model to Reduce Hallucinations

Model selection directly influences hallucination risk and output quality.

• Use higher-quality models for deep reasoning, strategy, and high-stakes work
• Use lighter, faster models for rewriting, brainstorming, and low-stakes tasks
• Use fact-enhanced or research modes when current information is required
• Prioritize reliability over speed when accuracy matters

Workflow & Productivity

Talk, Don't Type

Voice can be a useful way to give AI more context. Try dictation for detailed prompts and compare it with your own typing workflow.

Free Option

• Use the built-in dictation button in ChatGPT or Claude
• Works on desktop and mobile
• No additional tools required

Advanced Paid Option (Wispr Flow)

• Faster: save seconds on every prompt — thousands of interactions means hours saved per month
• Smarter: custom vocabulary and typo-correction libraries for your industry
• Snippets: voice-activated macros that recall your best prompts hands-free
• Cross-app: voice control across any application, not just your AI chat window

Building a Prompt Library

Once a prompt has been engineered, tested, and enhanced, save that final version so it can be recalled and reused.

• Preserve every high-performing engineered prompt — never lose a good one
• Organize by category and use case (e.g., Marketing, Legal, HR, Strategy)
• Update and refine prompts as AI models advance and improve
• Share your library with your team — it becomes a scalable organizational asset
• Use voice recall tools (Wispr Flow) to trigger prompts hands-free by keyword

Conclusion

Conclusion & Key Takeaways

• Clear prompts lead to clear results — vague input produces vague output
• Use the CREATE framework to structure every important prompt
• Provide strong, focused context — think of it as briefing a smart colleague
• Engineer your prompts before using them for high-value or repeated tasks
• Construct and maintain an organized prompt library for your team
• Superior prompts accelerate your workflow and meaningfully improve outcomes