Write prompts that get usable answers

What it is

Prompt engineering is the craft of writing instructions that steer a model toward the answer you actually need. Most weak output traces back to a weak request. A model is not a search box. It responds to direction, so treat each prompt like a brief you would hand a capable colleague who knows nothing about your situation yet.

A dependable structure has four parts. Give the model a role. State the task with a clear verb. Supply the context it cannot guess. Describe the format you want back. Skip any of the four and the model fills the gap with assumptions.

When to use it

Use this whenever a first attempt comes back generic, too long, or aimed at the wrong audience. It matters most for recurring work: reports, client emails, summaries, analyses that others will read. For a quick one-off question you can stay loose. For anything you will reuse or hand to someone else, spend the extra minute on structure.

How to use it

1. Assign a role. Name the perspective you want, such as a compliance reviewer or a copy editor for a trade magazine. This shapes vocabulary and priorities.
2. State the task with one strong verb. Create, compare, rewrite, audit. Avoid stacked requests like "look at this and maybe improve it".
3. Add context. Paste the relevant background, the audience, the constraints, and any prior decisions. This is where most prompts quietly fail.
4. Specify the format. Say the length, the shape, and the tone. A table with three columns beats "make it clear".
5. Choose a technique. For a routine job, a single clear prompt is enough. Show one or two examples when you need a specific style. Ask the model to sketch the wider picture before drilling into detail when the problem is unfamiliar. Ask it to reason in stages when the answer needs logic.
6. Match the model to the job. Fast models suit summarising and rewriting. Slower reasoning models suit multi-step planning. Reserve the deepest research modes for decisions where being wrong is expensive.
7. Review and revise. If the output misses, do not start over. Point at what was wrong and ask for a corrected version.

What a good result looks like

Here is a worked example. A facilities team wants to cut energy costs. A weak prompt is "give me ideas to save energy". A structured one reads: "You are an operations consultant. Draft a costed action plan to cut electricity use in a 40-person office over six months. Context: the lease ends next year, the landlord shares metering, and staff resist temperature changes. Format: a table with action, cost, monthly saving, and owner, then a short paragraph on the first step." Ask for reasoning in stages if the trade-offs are complex.

The difference shows immediately. The second prompt returns something a manager can act on: named actions, numbers, accountable owners, and a starting point. Clear input, specific output, less rework.