Prompting Fundamentals

Prompting is the closest thing AI builders have to a universal skill — it matters whether you're chatting casually or building a product on top of a model's API. And despite the volume of "secret prompt tricks" content online, the fundamentals are simple and stable.

The four things a strong prompt specifies

• Context — background information the model needs but doesn't already have.
• Task — the specific thing you want done, stated as directly as possible.
• Format — how you want the output structured (a list, a table, a specific length).
• Constraints — anything the output must avoid or must include.

Most disappointing AI outputs trace back to one of these four being missing, not to any deep limitation of the model. "Write a marketing email" is missing almost all four. "Write a 150-word marketing email for a bakery's new sourdough loaf, in a warm and casual tone, ending with a call to visit this weekend" is a completely different — and far more useful — prompt.

Show, don't just tell

When you can, include an example of the kind of output you're after. Models are very good at matching a pattern once they've seen one, often better than they are at interpreting an abstract description of style or structure. This is especially powerful for tone, formatting, and structured data.

Treat your first prompt as a draft, not a final answer — the best results usually come from a short back-and-forth.

Iterate instead of over-engineering

New builders often try to write one perfect, exhaustive prompt up front. In practice, it's usually faster to send a reasonable prompt, look at what comes back, and refine: "shorter," "more specific," "remove the third point." Conversational iteration is a feature of these tools, not a workaround — use it.