Write clearer AI prompts and get more useful drafts

1. Lead with the outcome you want

“Tell me about marketing” leaves room for very different responses. Naming the task and audience narrows the request: “Draft a product introduction email for a small shop.” You can then specify the tone, length and information the draft should include.

Break independent outputs into stages. Review an outline first, then ask for the text of each section. This lets you correct the direction before the tool produces a long response that is difficult to review or reuse.

2. Explain the context and information boundaries

Describe the audience, the intended use and the information the model may rely on. If the response must use only a supplied document, say so explicitly. Ask it to identify missing information and request clarification rather than fill gaps with guesses.

An example or reference is useful when it directly affects the result. Do not upload a complete customer file or confidential material simply to provide more context. A short, anonymized sample will often explain the structure of the task.

3. Define the format and acceptance criteria

“Keep it short and good” is difficult to evaluate. Specify the number of sections, approximate length and type of output: three key points, a comparison table or a short explanation for a non-specialist. A small example of the desired format can remove further ambiguity.

Separate accuracy requirements from writing style. A response can sound appropriate while changing a number or ignoring a constraint. Before using it, check both the factual content and whether it follows the requested structure.

Use the source text below to prepare a project update.
Output three sections: Decisions, Next actions, and Open questions.
For each action, include an owner and deadline only if the text gives them.
Do not invent names or dates. Write “not specified” where information is missing.
Source text:
[Insert a version without sensitive information.]

4. Ask for a specific revision

After the first response, name the part that should change and explain why. “Improve it” gives little direction. “Simplify the second section for a beginner and keep the example” is easier to assess. Changing one factor at a time makes the effect of a revision easier to see.

If the response has drifted from the goal, restate the main requirement. Removing an unnecessary instruction or shortening the context can sometimes help more than adding another page of directions. Save a useful prompt and result as a reference for a similar task.

5. Review even a promising response

A clear prompt does not replace human review. Compare important names, figures, quotations and claims with the input and original sources. If someone else will use the output, inspect it in its final setting too; a long table may be awkward to read on a phone.

For recurring tasks, make a template with clearly marked variables. Keep the audience, source, constraints and output format separate. When the task changes, these fields help prevent an old instruction or outdated data from carrying over unnoticed.

Your next-step checklist

Review each item before using the result.

Further reading

This guide suggests a workflow. Check the official source and the specific product page for current features and terms.

Related tools in the ElmosGPT catalog

ChatGPTClaudeGemini