A03: How to Ask: Prompting

The single biggest difference between people who get great results from AI and people who get junk is not the tool, it is how they ask. A prompt is a brief. A vague brief gets vague work. This lesson makes your briefs sharp.

Vague In, Vague Out

Compare:

• Weak: "Write about dogs."
• Strong: "Write three short tips for a first-time dog owner on house-training a puppy, one sentence each, plain language."

The second one says the topic, the audience, the format, and the length. The model no longer has to guess, so it stops guessing and starts helping.

The Four Levers

You control four things in almost every prompt:

• Task - what exactly do you want done? "Summarize," "compare," "fix," "explain."
• Audience / level - "explain to a complete beginner," "assume I know nothing about code."
• Format - a list, a table, numbered steps, one line. Ask for it and you get it.
• Constraints - length, language, tone, what to avoid. "Under 100 words." "No jargon."

Add an example of what good looks like when you can. Showing beats describing.

The First Answer Is a Draft

You are not stuck with what comes back. Push on it:

• "Shorter." / "Give me a concrete example." / "You assumed I use Windows, I use a Mac."
• "That link, does it actually exist? Show me the source."

And avoid the trap from A01: do not ask "is this good?", the model is built to say yes. Ask "what are three problems with this?" You get far more by inviting criticism than by fishing for approval.

This Week's Exercise

1. Take a real question you have this week. Write the laziest one-line version of it.
2. Rewrite it three ways: once adding format, once adding audience/level, once adding constraints.
3. Run all four prompts. Note how the answer changed each time.
4. Pick the best answer, then push on it once more ("shorter" / "with an example" / "what's wrong with this?"). Bring the before-and-after to class.