🧬 The 5 parts of a strong prompt
Every professional prompt, in any tool and for any task, is built from the same 5 pieces: Role, Task, Context, Format, and Example. You don't always need to use all 5—but you need to know which one is missing when the answer is bad.
THE 5 PARTS IN A REAL PROMPT (the course's universal template)
Your task is [TASK]. ← clear verb + clear object
Context: [WHAT IT NEEDS TO KNOW]. ← situation, audience, constraints
Format: [LIST / TABLE / STEPS]. ← what the response should look like
Example of what I want: [1 EXAMPLE]. ← one template is worth a thousand instructions
💡 Practical Tip
When an answer is bad, don't rewrite everything — diagnose which of the 5 parts was missing. Generic answer? Context was missing. Wrong tone? Role was missing. Messy? Format was missing. Diagnosis is faster than starting over.
⚔️ The difficult email — the same request in 4 versions
The situation: a client bought your product, had a delivery problem, and sent an angry message. You need to respond today. See what happens with each version of the request—this progression is the heart of the entire track.
Version 1 — the raw request
What you get: a polite, completely generic text — “Dear customer, we regret what happened...” It could be for any company, any problem, any planet. You wouldn’t use it.
Version 2 — task + situation (basic context is added)
What you get: visible improvement—the response mentions the delay and the packaging. But the tone sounds like a corporate robot and offers nothing concrete. Usable with a lot of editing.
Version 3 — role, tone, and what to offer are added
What you get: a human-sounding response that owns the problem and offers a concrete solution. It would already be ready to send with minimal edits.
Version 4 — all 5 parts complete (format + example)
What you get: an email ready to send, in YOUR tone, the right length, with a thoughtful emotional structure. From “reply to this customer” to this: about 60 extra seconds of typing.
✗ What a weak request does
- ✗Outsources ALL decisions to AI (tone, length, offer)
- ✗Generates an answer that works for everyone = works for no one
- ✗It forces you to rewrite—the “time savings” turn into a loss
✓ What a strong request does
- ✓You make the important decisions; AI executes exceptionally well
- ✓Answer in your tone, for your situation and your client
- ✓60 more seconds on the request = 20 fewer minutes of editing
Progression
V1→V4: each added part visibly improves it
Your decisions
Tone, offer, and structure: you decide
Cost × return
1 min of prompting saves 20 min of editing
Reuse
V4 becomes a template in module 1.2
📐 Output format — ask for the answer in the format you’ll use
Those who don’t ask for a format get whatever format AI chooses—usually long paragraphs. The right question to ask before making a request: "where will this answer end up?" In a spreadsheet? Ask for a table. For a checklist? Ask for steps. For a slide? Ask for 5-word bullets.
FORMAT PHRASES TO ADD TO THE END OF ANY PROMPT
"List them in numbered steps, 8 steps max."
"Summarize in 5 bullets of no more than 10 words each."
"Prose, maximum 150 words, 2 paragraphs."
"Start with a 2-line summary, then the details."
"End with a question that helps me decide on the next step."
📊 Why this matters so much
- Format = usable response without rework. The requested table is ready to paste directly into the spreadsheet.
- Format limits length. "Max. 150 words" nips verbosity in the bud.
- Format forces clarity. Asking for it "in steps" forces the AI to organize its reasoning in sequence.
💡 Practical Tip
Forgot to ask for a format? No worries: ask for reformatting in the next turn — “turn this into a table,” “now in 5 bullets.” AI reformats its own work in seconds. You saw this in the first guided conversation in module 1.0.
🎚️ The right level of explanation
The same content can be explained to a child, a layperson, or an expert—and AI does all three perfectly, as long as you say which one you want. See the same topic (compound interest) at 3 levels:
LEVEL 1 — CHILD (to help YOU understand something from scratch)
LEVEL 2 — INTERESTED BEGINNER (to create accessible content)
LEVEL 3 — EXPERT (to work as an equal)
💡 Practical Tip — the superpower of joining late
The “child” level is the the world’s best private tutor for yourself. Any term in AI, finance, or technology that intimidates you: "explain it like I'm 10 years old." Then move up to level 2. In two questions, you'll understand what seemed inaccessible—that's how a latecomer quickly becomes a veteran.
🧪 Examples and intentional comparisons
The most powerful instruction there is isn’t an instruction—it’s a example. AI imitates style, structure, and tone with impressive accuracy. Instead of describing what you want in 3 paragraphs, show 1 sample.
TECHNIQUE 1 — STYLE EXAMPLE ("write like me")
TECHNIQUE 2 — GUIDED COMPARISON (to make better decisions)
🔍 Why examples work so well
Remember module 1.0: AI is a pattern machine. An example is a concrete standard — much easier to follow than abstract adjectives like “professional but casual.”
Practical rule: 1 good example > 3 paragraphs of description. Examples of what NOT to do work too ("avoid this style: ...").
💀 The 3 mistakes that kill a prompt
90% of bad answers aren't the AI's fault — they're caused by one of these 3 killers in your request. Learn to spot them in the prompt itself before pressing enter.
✗ The killer
- ✗Generic: "Talk about digital marketing" — no audience, no objective, no focus
- ✗Ambiguous: "Summarize the document and the important points from the meeting" — summarize what, exactly? Both? Together?
- ✗No format: "Give me a few post ideas" — how many? For which platform? With a caption or just a concept?
✓ The antidote
- ✓Scope: "3 digital marketing strategies for a neighborhood bakery with a R$300/month budget"
- ✓One request per sentence: "1) Summarize the document in 5 bullets. 2) Then list the meeting decisions."
- ✓Specification: "5 Instagram post ideas, each with a 1-line hook + image suggestion"
⚠️ The intern test
Before you send, ask: "if I gave this request to a smart intern on their first day, would they know exactly what to deliver?" If the answer is no, the AI doesn’t know either. Fix the request, not the response.
Generic
Antidote: narrow the scope (audience + goal + limit)
Ambiguous
One request per sentence, numbered
No format
Antidote: quantity + structure + destination
Intern test
A clear request for a person is clear for AI
🔁 Iteration — improve the response without starting over
The classic beginner’s mistake: the response is 80% good, so they delete everything and rewrite the prompt from scratch. A fluent user does it differently: uses the 80% and adjusts the 20% in the next turn. This is cumulative—use it.
REFINEMENT COMMANDS — the vocabulary for the second round
"Good structure, but the tone is too formal — make it more casual."
"Replace example 3 with one from [FIELD]."
"Now the same thing, in a version for WhatsApp (3 lines)."
"Give me 3 variations of the title only."
"What did you assume that I didn't say? List it before continuing."
📊 The math of iteration
A reasonable prompt + 2 well-chosen tweaks gets you to the result faster than 10 attempts at a “perfect prompt” from scratch — because each adjustment builds on what’s already good. Perfectionism in the first prompt is beginner procrastination in disguise.
⚡ Slow User Shortcut
Memorize ONE adjustment command: "keep everything, just change X". It's the Swiss Army knife of iteration—it preserves what's good and surgically replaces what isn't. With this one habit, you'll already iterate better than most people.
🧬 Module Summary
📦 Module deliverable:
Take 3 real prompts you’ve already used (or would use) and rewrite each one with the 5 parts. Test the before and after in the tool. Save the 3 “after” versions—they’ll be the first items in your library in module 1.2.
Next Module:
1.2 — Reusable prompts: from improvisation to a template (and 10 ready-to-use templates)