Six prompt ingredients that cut ambiguity

A prompt that works for a real task is more than a vague request. It states a clear goal—what you want the AI to achieve—and often assigns a role or audience, like 'you are a policy analyst writing for non-experts'. Constraints, such as word limit or tone, stop the output from drifting. Source material gives the AI specific facts to use, and output format tells it whether you need a table, bullet list, or paragraph. Finally, verification instructions ask the AI to check its own work, for example 'list any assumptions you made'. Without these six parts, the AI guesses, and you waste time editing.

To spot a weak prompt in a question, look for missing constraints or an undefined output format. If the task says 'summarise this report' but does not give a length or audience, the answer is incomplete. A useful trick: imagine you are giving instructions to a new colleague who follows orders literally. If they would ask 'how long?' or 'who for?', the prompt needs those details. Another hint: verification instructions are not just 'check for errors'—they can be 'cite the source for each claim' or 'flag any data older than 2020'. This step is often overlooked in exam scenarios.

To test yourself, take any work task and write a prompt using the six parts in order: goal, role, constraints, source, format, verification. Then remove one part and see how the output changes. The memory anchor is 'GRaC-SoFV'—goal, role, constraints, source, format, verification. If a prompt in an exam misses any of these, it is likely the wrong answer.