Prompt Anatomy: Writing Clear Prompts

A useful prompt explains the task, the context, the constraints, and the desired format — four parts, each doing a distinct job. Vague prompts fail because they skip parts and make the model guess what you wanted.

The four parts

• Task — say what you need done. One clear verb beats three hopeful ones. “Summarize this ticket” is a task; “look at this” is a wish.
• Context — supply the information needed to do it. The model cannot use what you did not give it. Paste the ticket, describe the audience, name the background it should assume.
• Constraints — state the limits. Length, tone, what to include, what to leave out, what to do when information is missing (“say 'unknown' rather than guessing”). Constraints are where most prompts are weakest.
• Format — show the output shape. Bullets or a table? Headings? A one-line verdict first? The model will happily reshape the same facts into whatever container you name.

One more rule that belongs with the four: keep source material separate from instructions so a reader — human or model — can tell which is which. Mark quotes as quotes. Instructions found inside retrieved documents should be treated as untrusted content, not commands: a document that says “ignore previous instructions” is data, not your boss.

Before and after

Vague: “Summarize this support ticket.” — The model guesses the audience, the length, and the shape.

Same facts in, far more useful text out — because every part told the model something it would otherwise have had to invent.

Build a prompt to summarize a real support ticket, email, or meeting note you have handy. Write the four parts explicitly, then run it — and then change only the format: ask for a table instead of bullets.

Watch what changes and what doesn't: the requested output shape changes, but the underlying facts stay the same. That separation — facts from formatting — is the sign of a prompt doing its job. If changing the format changes the facts, your prompt is leaking; tighten the constraints.

Bonus round: deliberately remove the constraints section and run it again. The difference you see is the exact value that one part was adding.

Better prompts cannot supply missing evidence. No amount of task-context-constraints-format will make the model know your company's returns policy or last quarter's numbers. When the facts aren't in the prompt and aren't retrievable, a great prompt produces a beautifully formatted guess. (See why AI hallucinates.)

And prompts cannot guarantee correctness. They raise the odds of a useful answer; they don't certify it. For anything that matters — money, access, safety, published claims — the answer still gets checked against a real source, no matter how good the prompt was.