Basics

Prompt basics

The anatomy of a prompt that gets a usable answer.

Knowledge cards
Knowledge card
Put the Instruction Before the Data
A prompt-structuring habit where you state the instruction first, then paste the long data — the document, code, or transcript — clearly separated below it, rather than burying the ask at the end of a wall of text.
L1 · Basic prompts
Best practice
Show Claude One Example of What Good Looks Like
Attaching a single concrete example of the output you want — the right format, tone, and depth — instead of describing it in the abstract.
L1 · Basic prompts
Knowledge card
Tell Claude the Format You Want Back
A core prompting habit of telling Claude the shape you want the answer in — a table, a bulleted list, JSON, a two-sentence summary, a specific set of headings — rather than leaving the format to chance and reworking it after.
L1 · Basic prompts
Knowledge card
Say What to Do, Not Just What to Avoid
A prompting habit of phrasing instructions as what you want Claude to do, not only what to avoid — 'write in plain, warm English' rather than 'don't be so formal and don't use jargon'.
L1 · Basic prompts
Claude prompt basics

The anatomy of a usable prompt

A prompt that gets a usable answer has four parts, and most weak prompts are missing one. Role: who Claude should be. Context: the situation and any constraints. Task: the specific outcome you want. Standard: what a good answer looks like, ideally with an example. You do not need all four every time, but the more of them you supply, the less Claude has to guess. Guessing is where generic or off-target answers come from — not from the model being weak, but from the prompt being thin.

Lead with the outcome

A subtle but powerful habit is to describe the destination rather than dictate the route. 'Write a function that does X, handling these edge cases' lets Claude find a good implementation; a step-by-step recipe often boxes it into a worse one. State the outcome and the constraints, then let Claude reason toward it. Save the detailed steps for when you genuinely need a specific method — otherwise you are doing the thinking Claude is good at, and doing it less well.

Frequently asked
What makes a good Claude prompt?

Four elements: a role for Claude, the context and constraints, the specific task, and the standard for a good answer. The more you supply, the less Claude has to guess.

Should I tell Claude the exact steps to follow?

Usually no — describe the outcome and constraints and let Claude find the route. Detailed steps are for when you need a specific method; otherwise they can box it into a worse solution.

Why does Claude give me off-target answers?

Almost always because the prompt was missing a role, context, a clear task, or a standard, forcing Claude to guess. Supplying those removes the guesswork.