Show Claude One Example of What Good Looks Like
What it is
Attaching a single concrete example of the output you want — the right format, tone, and depth — instead of describing it in the abstract.
Why it works
Adjectives like 'concise' or 'professional' mean different things to you and to Claude. One example collapses that ambiguity: the model can copy a shape far more reliably than it can guess an intention.
When to use it
Whenever output format or style matters and you can produce or find one good sample — table layouts, commit messages, email tone, structured data.
When not to use it
Open-ended, exploratory prompts where a single example would narrow Claude's thinking too early.
Prompt
I want output in this exact shape. Here is one example:
<a single well-formed example>
Now produce the same for: <your real input>.Example
Given one sample changelog entry, Claude formats twenty raw commits into matching entries — same tense, same length, same emoji-free style — with no further instruction.
Advanced version
If the first output drifts, add a second example that specifically covers the edge case it got wrong. Two well-chosen examples usually pin the behaviour completely.
Common mistakes
- Describing the format in words when a five-line sample would have been unambiguous.
- Giving an example that is subtly inconsistent, so Claude faithfully copies your mistake.
- Over-exampling an open question and boxing the model into your first idea.