How to Write Better AI Prompts (Plus a Free Library to Start From)

Two people can ask the same AI tool about the same topic and get wildly different quality answers — not because one has a "better" AI, but because one gave it enough to work with and the other didn't. Prompting isn't a magic skill; it's just being specific about what you actually want.

What actually changes the output

• Role — telling it who to act as ("an experienced copywriter," "a senior backend engineer") shifts vocabulary and focus, not just tone.
• Audience — "explain this to a beginner" vs "explain this to another engineer" produces genuinely different depth, not just different words.
• Format — asking for a specific output shape (bullet list, table, exact word count) stops you from having to reformat the answer yourself afterward.
• Constraints — explicit limits ("under 100 words," "no jargon," "don't include X") do more work than most people expect, because without them the model guesses at your preferences.

A concrete before/after

• Vague: "Write about my product launch."
• Specific: "Write a 100-word announcement for a new invoicing feature, aimed at small business owners, in a friendly but professional tone, ending with a clear call to action."

The second version tells the model the length, audience, tone and what the ending should do — four decisions it would otherwise have to guess, and probably guess wrong on at least one.

Step-by-step: write a better prompt

1. State the role you want the AI to take, if relevant to the task.
2. Say who the output is for.
3. Specify the format and length you actually need.
4. Add any hard constraints (things to avoid, exact numbers, must-include details).
5. If the first answer is close but not right, don't start over — tell it specifically what to change.

Common mistakes

• Assuming a bad first answer means the tool "doesn't understand" the topic, when it usually just didn't have enough context — adding detail almost always helps more than switching tools.
• Not giving feedback on a near-miss answer — "make it shorter" or "more casual" is a completely valid follow-up, not a failure.
• Over-specifying a creative task so tightly that there's no room left for a genuinely good idea — constraints help most for factual or structured output, less for open-ended brainstorming.