Have Claude Restate the Task in Its Own Words

A beginner move where, for anything non-trivial, you ask Claude to restate the task and its assumptions in its own words before it starts — so you can catch a misunderstanding while it's still free to fix.

Most bad answers aren't a reasoning failure; they're an answer to a slightly different question than you meant. Once Claude has produced a long, confident response to the wrong task, correcting it costs a full round trip. A restatement surfaces the misread up front, when fixing it is a one-line correction. It also exposes hidden assumptions you didn't know you'd left ambiguous.

Multi-step tasks, anything with ambiguity, and any prompt where a wrong direction is expensive to redo — a long document, a big code change, a plan you'll act on.

Simple, unambiguous requests where the restatement is slower than just doing the thing. Don't make Claude echo 'translate this sentence' back to you.

Before you start, restate what I'm asking for in your own words, list the assumptions you're making, and note anything ambiguous you had to guess about. Don't do the task yet — wait for me to confirm or correct.

You ask for 'a summary of this report for the board'. Claude restates: 'a one-page, decision-focused summary for non-technical directors, emphasising financial impact'. You realise you actually wanted the technical risks foregrounded, correct one word, and get the right summary the first time.

Ask Claude to also state what it will not do — the things it's treating as out of scope. The out-of-scope list catches the opposite failure: not a misread of what you wanted, but a silent omission of something you assumed was included.

• Skipping the confirmation and letting Claude run with its restatement, which wastes the step.
• Writing a prompt so vague that even the restatement can't pin it down — give it something real to reflect back.
• Using it on trivial tasks, where it just adds a turn of ceremony.