AI & Productivity

How do I write better AI prompts?

Use a repeatable prompt structure

A prompt is an instruction and the context around it. Clear prompts reduce guesswork, but no template can guarantee a correct or useful answer.

Try four parts: task (what to do), context (what the system needs to know), constraints (what to include or avoid), and format (how to present the result). Include examples when the style or structure matters.

  • Task: explain a topic, summarize notes, or compare options.
  • Context: audience, purpose, and source material you may share.
  • Constraints: length, reading level, boundaries, and exclusions.
  • Format: steps, table, checklist, outline, or plain-language paragraph.

Make a prompt specific without overloading it

Replace broad requests such as 'tell me about design' with a concrete outcome, such as 'give me a beginner checklist for reviewing contrast and text size in a one-page poster.' This tells the tool what a helpful answer should contain.

Provide only the context needed for the task. Split a complex job into stages, inspect each result, and correct errors before moving to the next step. Long prompts can still be ambiguous if the goal is unclear.

Review the response

Check whether the response follows the requested format, covers the important parts, and makes claims you can verify. Ask a follow-up to clarify or revise, but do not treat repeated agreement as independent evidence.

For a reusable workflow, save a prompt only after testing it on varied examples. Note where it fails and keep a human review step for anything published, submitted, or used to make decisions.

Tools mentioned

ChatGPT, Gemini, Claude. Features and availability can change; check each provider's official information before choosing a tool.

Frequently asked questions

Do longer prompts always work better?

No. Relevant context and a clear task matter more than length. Remove details that do not help the system produce the requested result.

Can a prompt prevent AI errors?

A clear prompt can reduce ambiguity, but it cannot guarantee factual accuracy. Review and verify important output.

What should a prompt include?

The role if it helps, the task, the context, and the format of the answer. Leave out stories that do not change the result.

Should I say "act as an expert"?

Only if you also say what expert behavior you need. The label alone does not add facts.

How do I ask for a revision?

Quote the weak part and say what to change. "Make it better" is too vague.

Can I save prompts?

Yes. Keep the ones that worked, next to an example of the good output, and delete the ones that only sounded clever.