Prompt Design
Prompt Design for Generative AI
The Golden Rule: Treat the AI Like a Smart New Colleague—Every Time
In LearningFlow, every time you use an AI-powered node, you’re starting with a fresh, talented helper—like a highly-educated intern or new graduate, eager to please but unaware of your ongoing project unless you tell them.
They know science, math, and grammar—but not your goals, class, or yesterday’s lesson!
Always brief the AI completely: give audience, topic, goals, format, and anything a “new team member” would need to succeed.
Principles of Effective Prompting
1. Be Specific and Clear
- Avoid:
Write questions.
- Better:
Write 5 short-answer questions about plant photosynthesis for grade 6, each with model answers.
2. Give Context and Role
- Pretend the AI knows nothing about you or your students.
- Example:
You are an English teacher planning for a class of 14-year-old intermediate learners in Singapore. Prepare a grammar exercise on past tense verbs.
3. Request Output Format
- Tell the AI how you want the information presented.
- Example:
List the answers at the end in a separate answer key section.
Give your answer as a table with two columns: Problem | Solution.
4. Use Examples If Helpful
- Show the AI what you mean.
- Example:
Sample: "What is the capital of France? Paris." Now write 3 similar questions for Asian capitals.
5. Break Down Complex Tasks ("Chain of Thought" or Multi-Step)
- Guide the AI through each required step.
- Example:
First, explain Pythagoras’ Theorem in simple terms. Then, provide one real-world word problem. Next, show the detailed solution step by step.
Common Prompt Pitfalls (and How to Fix Them)
Pitfall | Example | How to Improve |
---|---|---|
Too vague | Create a worksheet. | Add topic, number, level, format |
No audience info | Write questions. | Specify class age, level, subject |
Missing output structure | Summarize renewable energy. | Summarize in 3 bullet points for teens. |
Too much packed into 1 ask | Long paragraph prompt | Use bullets, clear steps (“First..., then...”) |
Assuming knowledge/context | Continue from previous message. | Restate the context in every prompt |
Output Formats — Be Explicit
- List:
List 8 key terms, each on a new line.
- Table:
Present the advantages and disadvantages in a table.
- Dialogue:
Write a conversation between a customer and a restaurant server, 4 exchanges each.
- Multiple choice:
Each question should have 4 options and indicate the correct one.
Best Practice Checklist
- Who is your audience? (age, background, learning goals)
- What is the subject/topic and level of difficulty?
- Is a specific tone or role required? (“Friendly tutor”, “strict examiner”, etc.)
- Have you requested a format for the output?
- Did you provide any content or style examples?
- Have you broken down complex asks into steps?
Iteration: Improve Through Testing
- Don’t settle for the first result!
Tweak and reword your prompt—try again. - Save your best prompts for future use.
- Compare outputs with and without audience/context/format included.
Ready to Explore Further?
Here are different prompt styles and patterns you can try. Click any to see detailed examples and guidance (create these pages as you grow your documentation):
- Single-Instruction Prompts
- Role-Based Prompts
- Multi-Shot Prompts (with multiple examples)
- Chain of Thought Prompts
- Output Format Prompts
- Dialogue and Scenario-Based Prompts
- Prompting for Branching/Scenarios
- Prompt Debugging Examples
- Creative Prompts (Story/Art/Brainstorming)
Practice Makes Perfect
Test these prompt styles in your own flows using LearningFlow’s AI nodes.
The more you experiment, the faster you’ll develop a “feel” for what gets great results for your learners.
Next:
Learn how to evaluate AI results for accuracy, appropriateness, and depth.
Every flow begins with a prompt. Make every prompt count—your AI “colleague” is ready to help, if you guide them well!