Open With the Outcome You Want, Then the Details

A simple prompt structure for beginners: lead with the outcome you actually want — the finished thing and who it's for — then give constraints and context, and end with the specific ask. The goal comes first because everything else is in service of it.

When the outcome is buried or implied, Claude optimises for the words it can see rather than the result you need, and you get a technically-responsive answer that's useless for your actual purpose. Stating the destination first lets the model shape everything — length, tone, depth, format — toward it, and it forces you to know what you want, which is half the battle.

Every prompt where the output has a job to do — a document for a specific reader, code for a specific system, an answer that feeds a decision. Especially when you're new and unsure how much to say.

Genuinely open exploration where you want to be surprised, or trivial factual questions where there's no 'outcome' to speak of.

I want <the outcome: what, for whom, to do what>.
Constraints: <length, tone, must-haves, must-avoids>.
Context you need: <the relevant facts>.
Now: <the specific ask>.

'I want a rejection email that keeps the candidate warm for future roles (outcome). Two short paragraphs, warm but honest, no false promises (constraints). They interviewed well but we picked someone with more infra experience (context). Draft it (ask).' — a far better result than 'write a rejection email'.

When you're not sure what good looks like, add one line of 'success criteria' — how you'll judge the output — and ask Claude to meet them. Naming the criteria often improves your own request as much as the answer.

• Leading with a pile of context and never stating the outcome, so Claude guesses the goal.
• Confusing the task ('write an email') with the outcome ('keep them warm for later').
• Over-specifying tiny details while leaving the actual purpose unsaid.