Prompting
Prompt optimization
Systematically improving prompts based on measured output quality.
Example
Repeated evaluations reveal which instruction changes improve accuracy without adding unnecessary text.
Why it matters
It reduces wasted attempts and helps teams find instructions that work reliably for their needs.
You'll hear it
“That revision helped ordinary cases but made exceptions worse.”
Can you control it?
Yes
Yes. You can often use or adjust prompt optimization directly when the product supports it. The available options and their effect still depend on the tool.
Common questions
- Can a prompt become too tailored to its test examples?
- Yes. It can perform well on familiar checks without improving on new work.
- Does a shorter prompt always cost less overall?
- Not necessarily. Extra retries or poorer results can offset the saving on each request.
- Can automated optimization choose bad shortcuts?
- Yes. It may favor whatever improves the chosen score, even if that misses the user's real goal.