Six elements, sharper responses
Most AI outputs are only as good as the prompt that produced them. A vague prompt produces a vague response. A well-structured prompt gives the AI enough to work with and enough constraint to stay on target.
Six elements provide a reliable structure for constructing prompts that get useful responses. You do not need all six every time — a quick factual question needs none of them, and a complex analytical task benefits from most of them. The skill is knowing which elements matter most for the task you are doing.
| Element | What it does | Example |
|---|---|---|
| Role | Shapes the lens the AI responds through. Telling the AI who to be changes how it frames its answer — a sceptic, a specialist, or a critical friend all produce materially different responses to the same question. | "You are an executive coach with expertise in conflict resolution." |
| Context | Provides the background, goals, or challenge you are navigating. The AI has no access to your situation unless you provide it. More context produces more calibrated responses. | "I am preparing for a meeting with two colleagues who disagree over project priorities." |
| Request | States precisely what you need. Ambiguous requests produce ambiguous responses. Specificity about what you want — and what you do not want — significantly improves output quality. | "Give me a five-step plan for opening the meeting and closing with an agreed way forward." |
| Constraints | Sets the boundaries for the output. Length, tone, perspective, what to exclude. Constraints save significant reshaping time and prevent the AI from producing something technically correct but practically unusable. | "Under 400 words, professional tone, UK English." |
| Format | Defines how you want the response structured. Saves time reformatting an answer that would otherwise need significant reworking before it is usable. | "Numbered list with a short explanation for each step." |
| Confirmation | Asks the AI to flag anything missing or unclear before it starts. Catches gaps in the brief that would otherwise produce a well-written answer to the wrong question. | "If I have missed something important, tell me before you answer." |
Treat the AI as a thinking partner, not an answer machine. A good first response is a starting point, not a finished output.
Two pitfalls worth knowing
The first is jumping straight to the request without setting role or context. This produces generic responses that feel plausible but do not engage your actual situation. The AI is working blind on what matters to you and why.
The second is treating the first response as final. Iteration — asking the AI to simplify, expand, reframe, or challenge its own answer — consistently produces better outputs than a single well-crafted prompt. The exchange is a conversation, not a transaction.
Using the framework well
Start with Role and Context
These two elements do the most work. Setting a clear role and grounding the request in real context consistently produces more useful responses than any other single change.
Use iteration
A good first response is a starting point, not a finished output. Ask the AI to simplify, expand, or reframe. Treat the exchange as a conversation, not a one-shot transaction.
Switch roles to unlock perspectives
Asking the AI to respond as a sceptic, a regulator, or a customer who disagrees can surface considerations that a single framing misses entirely.
Ask for a confidence check
Adding "let me know how confident you are in this response" or "flag anything you are uncertain about" surfaces the limits of what the AI actually knows, rather than leaving them hidden behind fluent prose.
Calibrate to the task
Not every prompt needs all six elements. A quick factual query needs none. A complex analytical task needs most. The skill is knowing which elements carry the most weight for what you are trying to do.
Bring your own knowledge in
The AI has no access to your organisation, your context, or your professional judgement unless you provide it. The more you put in, the more calibrated the output.