Blog
5 min read

How to write prompts that actually work

Most bad AI output is a briefing problem, not a model problem. A good prompt reads like instructions you'd give a competent stranger.

Treat it like briefing a person

The fastest way to improve a prompt is to imagine handing it to a capable stranger with no context. Would they know what you want, in what format, avoiding what? If not, the model won't either. Vague in, vague out.

A structure that holds up

  • Role and goal: who the model should act as, and what success looks like.
  • Context: the background it needs but wouldn't otherwise have.
  • Constraints: length, tone, format, and what to avoid.
  • An example: one sample of the output you want, if you can give one.

The example does the most work. Showing one good output usually beats three paragraphs describing it.

Keeping the ones that work

Once a prompt reliably produces what you want, it's an asset — and most people lose it in a chat history. Bizwax's PromptLine is built for exactly this: writing, testing and saving the prompts that work so you can reuse them instead of rewriting them from memory each time.

PromptingPrompt engineeringAI

Try it on your own keys

Bizwax.ai is in closed beta. Request access and we'll email you when you're in.

Request access

Keep reading