Most weak prompts contain the request and nothing else. The model then fills the gaps with the most typical answer it can produce, which is why the result reads as generic.
What makes a good prompt?
A good prompt reads like a clear briefing to a capable new colleague. Anthropic's prompting best practices make the same comparison, suggesting you treat the model as a brilliant but new employee who doesn't yet know your norms. A useful briefing covers:
- what you want done
- who it's for
- the material to work from
- how long it should be
- anything to avoid
The most useful single habit is to paste the thing rather than describe it. The actual email gets a better reply than your summary of the email.
None of this needs special phrasing or tricks. People who are good at briefing juniors tend to write good prompts, because it's the same skill.
Why does the prompt matter so much?
The quality of what comes back largely follows the quality of what goes in. A plain prompt with real material usually beats a clever prompt with none. That also means many disappointing results can be fixed by improving the prompt, before you think about changing tools.
A before-and-after example
Weak: "Write an email about the delay."
Stronger: "I run a small furniture workshop. A supplier delivered three weeks late and I want to complain without damaging the relationship. Write a short email, polite but clear that this can't happen again. Under 120 words."
The second version names the situation, the relationship, the tone and the length. What comes back needs editing, not rewriting.
Prompt vs system prompt
A prompt is what you type in the moment. A system prompt is a set of standing instructions loaded before the conversation starts, usually by the product, and it applies to every reply. Both take up room in the model's context window.
Common questions
Do longer prompts work better?
Only when the extra length adds useful context. Background the model needs improves the result, while padding and repeated instructions can make it worse. A good test is whether each sentence would change the output if you removed it.
Does the same prompt work on every model?
Roughly, but not identically. Models are trained differently and respond to phrasing in slightly different ways, so a prompt that works well on one may need small adjustments on another.
Should I save prompts I use often?
Yes. Keep the ones that work in a note so you can reuse them. In an AI agent, a prompt you give for the same job every week can become an agent skill, which the agent applies without you pasting it each time.