How to Write AI Prompts

Specificity. Vague prompts produce generic outputs — the AI has to guess what you actually want. The single biggest improvement most people can make is to add concrete details: the audience, the format, the length, the tone, and what to avoid. Even adding one specific constraint like "under 200 words" or "suitable for LinkedIn" measurably improves results.

Not necessarily. The best prompt is specific enough for the AI to produce exactly what you need — no longer. A 30-word prompt with clear role, task, and format beats a 300-word prompt full of vague instructions. That said, complex tasks like research reports or detailed analysis do benefit from more context. Start with the minimum and add detail only when the output is not specific enough.

Test it. A good prompt produces an output you can use with minimal editing on the first or second attempt. If you are consistently rewriting or regenerating, the prompt needs more specificity — usually more context, clearer format instructions, or explicit constraints. Keep a log of prompts that worked well and study what made them effective.

Core prompt structure works across all major AI tools. Role assignment, context, format specification, and constraints all improve results on ChatGPT, Claude, Gemini, and others. Minor differences: Claude tends to follow detailed constraint lists very precisely. ChatGPT responds especially well to role prompting. Gemini is strong for research and factual tasks. Start with one well-structured prompt and adjust based on the specific model's response patterns.

Do not start over — build on the response. Tell the AI specifically what to change: "Good structure, but make the tone more casual and reduce the length by half." Or: "The first paragraph is too formal — rewrite it to sound like a friendly email." Iterating is faster than re-prompting from scratch, and the AI retains your original context. Treat it as an editing conversation, not a one-shot request.