A prompt is the instruction you give to the AI. It's the element that has the most impact on the result: a good prompt is worth far more than any secret 'magic phrase'.

What should a good prompt contain?

Four ingredients: context (who you are, the situation), the precise task (exactly what you want), the expected output format (list, table, ready-to-send email…), and constraints (length, tone, what to avoid). Giving an example of the desired result helps enormously.

How do you improve your prompts?

By iterating: the first attempt is rarely perfect, you adjust by saying what doesn't work. Other habits: give an example, ask the AI to clarify before answering, have it play a role, and break a big request into steps.

Do you need to know secret prompt formulas?

No. There's no magic formula: clarity beats cleverness. A clear, contextualized, and precise prompt gives better results than any copied 'technique'.

How to talk to AI: writing a good prompt

1. What a prompt really is

A prompt is simply the instruction you write to the AI. Nothing technical about it. But it's the element that changes the result the most: the same AI can give you generic mush or a polished deliverable, depending on how you ask.

Forget the idea of a "secret formula" straight away. What works is clarity: saying precisely what you want, who it's for, and in what form. Like briefing a good intern — except they respond in ten seconds.

2. The ingredients of a good prompt

A solid prompt has four building blocks. You don't always need all four, but the more precise you are, the better:

ContextWho you are, the situation, who it's for. "I'm a hi-fi retailer, writing to a dissatisfied customer…"

TaskExactly what you want, in one action verb. "Write a reply", "summarize in 5 points", "compare these two options".

FormatThe output shape. Bullet list, table, ready-to-send email, max 3 paragraphs.

ConstraintsThe frame: length, tone, language, what to avoid. "Warm tone, no jargon, 120 words."

Context + task + format + constraints. Add an example of the desired output and you go up another level.

3. Bad vs. good prompt: an example

The same need, asked two ways. The difference in result is huge:

Vague

"Write me an email for a customer"

No context
No precise goal
Generic result, needs redoing

Precise

Context + task + format

"Customer unhappy about a delivery delay"
"Write a reply that apologizes and offers a gesture"
"Warm tone, 120 words, ready to send"

Same AI, same subject: the prompt makes all the difference.

4. The habits that change everything

Beyond the recipe, a few habits that save a lot of time:

Give an example. Showing a model of what you want beats ten lines of explanation.
Iterate without shame. "Too long", "more direct", "keep the 2nd point but shorten it": back-and-forth is normal, it's where things get good.
Ask it to clarify first before answering, when the subject is fuzzy. You avoid off-target responses.
Give it a role. "You're an SEO expert", "act as my CFO" usefully shapes both tone and substance.
Break it up. A big request split into steps gives better results than one monolithic block.

And when you find yourself typing the same instructions at the start of every conversation (tone, job, rules), it's time to lock them in once and for all in a file — that's exactly what the CLAUDE.md guide is about.