Tell the model what the output must accomplish, who will use it and what information it is allowed to rely on.
“Write something professional” sounds like an instruction, but it leaves almost every useful decision open. Professional for whom? A customer reading a refund message needs something different from an engineer reviewing a technical proposal.
A good prompt looks less like a secret formula and more like a short brief you would give a capable colleague.
Describe the job the output has to do
Begin with a concrete task. Instead of asking for “content about laptops,” ask for a comparison that helps a student understand the tradeoff between portability and repairability.
Name the audience and the decision. These two details make it easier to judge whether the answer succeeded. If the reader still cannot make the intended decision, impressive wording has not solved the problem.
Avoid stacking vague adjectives. “Premium, powerful, human, viral and world-class” adds little direction unless you explain what those words mean in this context.
Supply the facts that must stay fixed
Give the model relevant source material, product details or requirements. Separate facts from assumptions. If the information is incomplete, say which gaps should remain open.
Imagine a small store needs a return-policy summary. The useful input includes the actual return window, exceptions and refund method. Asking the model to invent a reasonable policy risks producing promises the store never agreed to honor.
You can instruct it to flag missing details before writing. That is often more useful than receiving a smooth draft filled with guesses.
Specify constraints that matter
State the output format, approximate length and any necessary exclusions. A two-paragraph email should not arrive as a ten-section article.
For example: “Explain this to a first-time customer in 150 to 200 words. Use the facts below. Include the next action and expected response time. Do not invent delivery dates.”
The constraints should support the task. An arbitrary demand for dozens of headings can make a short explanation harder to read rather than more useful.
Show one representative example
If tone or structure matters, include a short example and explain what you like about it. Perhaps it begins with the answer, uses ordinary language and avoids unnecessary apologies.
An example gives the model something concrete to imitate. Make clear whether it should follow the structure, voice or level of detail. Otherwise, it may copy features that were incidental.
Do not paste confidential examples into an unapproved tool. A fictional sample can communicate the same style.
Revise one weakness at a time
After the first answer, give specific feedback. “The opening is too long; put the required action first” is more actionable than “make it better.”
Check the factual content separately from the writing. A clearer prompt can improve relevance and organization, but it does not guarantee that every statement is true.
For repeated work, keep a small set of representative inputs and compare outputs after changing the prompt. This helps distinguish a real improvement from one unusually good result.
A reusable brief
- Task: What should the output accomplish?
- Audience: Who will read or use it?
- Facts: Which source material must it follow?
- Constraints: What length, format and boundaries matter?
- Success: What would make the result genuinely useful?
Fill these in plainly. You do not need theatrical roleplay or complicated terminology to communicate a well-defined task.
Sources & further reading
Original explainers and practical examples, with technical background from the sources below. Source links reviewed 2026-10-03.
More context, fewer assumptions. About our editorial approach.
