Your First Real Conversations
Master the basics of talking to AI effectively
Section 1: The Anatomy of a Good Prompt
The difference between a frustrating AI response and an excellent one usually comes down to how you ask the question. Let's break down what makes a prompt work.
The Four Essential Components
1. Context
Who are you? What's your situation? What domain are you in?
"I'm a marketing manager at a SaaS startup..."
2. Task
What specifically do you want AI to do?
"I need to write an email that..."
3. Format
How should the answer be structured?
"Give me bullet points..."
4. Constraints
What should you avoid? What's the length or tone?
"Keep it under 150 words, casual tone"
Bad Prompt vs Good Prompt
WEAK PROMPT
Problems: Too vague. No context. No target audience. No format specified. No length limit. AI has to guess.
STRONG PROMPT
Why it works: Clear context (who you are), specific task, target audience, format, length, and constraints. AI knows exactly what you want.
The pattern: "You are [role]. [Task]. The audience is [who]. Output as [format]. [Constraints]."
Section 2: The 5 Most Common Mistakes
These are the things that make AI responses disappointing. Fix them and you'll get dramatically better results.
Mistake 1: Being Too Vague
โ Vague
โ Better
Mistake 2: No Context About Who You Are
โ No context
โ Better
Mistake 3: Asking for Everything in One Prompt
โ Too much at once
โ Better
Mistake 4: Not Specifying Format
โ No format
โ Better
Mistake 5: Giving Up After One Bad Response
โ Giving up
"That wasn't good. AI is useless."
โ Better
The first response is rarely perfect. You're supposed to refine it. That's the normal workflow: ask, get a response, refine 1-2 times, use the result. Most people expect perfection on the first try and get discouraged. Don't be that person.
Section 3: Conversation Techniques
You can't just ask once and be done. Here are the key ways to refine and iterate with AI.
Technique 1: Ask for Alternatives
Don't settle for one response. Ask for variations.
When to use: When you're not sure which tone or style is right. Or when you want options.
Technique 2: Refinement
Tell AI what you want changed.
When to use: After the first response, when you see what you got but want tweaks.
Technique 3: Critique
Ask AI to evaluate its own work.
When to use: When you want to stress-test an idea or find holes before using it.
Technique 4: Stepping Up
Take the response and escalate its formality, depth, or impact.
When to use: When you've drafted something and now need to adapt it for a different audience.
Technique 5: Role Reversal
Ask AI to challenge your position.
When to use: Before presenting ideas to others. To prepare for objections.
Section 4: Hands-On โ Profession Walkthroughs
Here are six real, specific prompts for different professions. Pick yours and try it right now in Claude or ChatGPT.
The Task: Create a quiz that students will actually engage with.
What to look for: Are the questions clear? Are they the right difficulty? Would you actually use these?
Follow-up: "Make 3 of these questions harder for an advanced class."
The Task: Explain a diagnosis in a way patients understand.
What to look for: Is it understandable? Does it avoid scary language while being honest? Would a patient feel hopeful?
Follow-up: "Make this version for a teenager. Use language and examples they'd relate to."
The Task: Understand a complex topic for an exam.
What to look for: Can you understand this without being a history expert? Are the analogies helpful? Could you explain this to someone else now?
Follow-up: "What questions would a tricky teacher ask about this? Help me prepare."
The Task: Explain technical issues to non-technical people.
What to look for: Would a city councilor understand this? Does it explain the problem without using engineering jargon? Is it convincing?
Follow-up: "Now write a 30-second version I can say out loud in a meeting."
The Task: Get code review and feedback.
What to look for: Did AI correctly understand your code? Are the issues real? Are the suggestions practical or pedantic?
Follow-up: "Write test cases for this function, including edge cases."
The Task: Prepare for a client meeting.
What to look for: Is this actually useful for a real meeting? Does it give you talking points? Does it help you sound prepared?
Follow-up: "Draft an email to send after the meeting to recap next steps."
Section 5: The Refinement Loop In Action
Here's what a real conversation looks like. Three turns, each improving the result.
Turn 1: Your Initial Ask
Claude's Response (summary)
"Hi [Client], Thank you for thinking of us. Unfortunately, we're not able to take on the [Project] at this time due to capacity constraints. We'd love to revisit this in the future. Best regards, [Your name]"
Turn 2: Your Refinement
Claude's Response (revised)
"Hi [Client], Thanks so much for reaching out about [Project]. I appreciate your confidence in us. We're completely slammed right now, so I can't give it the attention it deserves. Let's grab coffee next monthโI'd love to explore what else we might do together. Talk soon!"
Turn 3: Final Polish
Claude's Response
"You could say 'tied up with other commitments' or 'at capacity' if you want it slightly less casual. But honestly, 'completely slammed' feels human and honestโclients know we're busy. It actually builds credibility. Keep it as-is."
The pattern: Ask โ get response โ tweak โ get better response โ final polish. Three turns, dramatically better result than the first attempt.
- "You are [role]. [Task]. [Context]. Output as [format]. [Constraints]."
- Context = who are you, what's your situation, what domain
- Task = exactly what you want AI to do
- Format = bullet points, essay, table, email, etc.
- Constraints = length, tone, what to avoid
- "Make it more concise"
- "Explain like I'm 5"
- "Give me 3 alternatives"
- "What are the limitations of this?"
- "Check this for errors"
- "Make it more formal/casual"
- "Add specific examples"
- "Summarize in one sentence"
- Ask for alternatives (3 versions with different tones)
- Refine (tell it exactly what to change)
- Critique (find weaknesses in the response)
- Step up (escalate formality/depth for different audience)
- Role reversal (ask for objections/tough questions)
- Being too vague = bad output
- No context about who you are = wrong tone/style
- Asking for everything at once = confusing response
- Not specifying format = whatever format AI chooses
- Giving up after first response = missed potential