Prompt workbench
Write a prompt for a small task, see what a real model produces from it, then get a critique of what your prompt did and did not say.
Sign in to try it. A free account includes a couple of runs.
Material the assistant can use
What the assistant produced
How your prompt did
| Element | Status | Note |
|---|
A stronger version
The critique did not come through this time. Your result above is still valid; try running it again to see the critique.
What to notice
- Vague prompts get generic output. The model fills the gaps with guesses, and the guesses are rarely yours.
- Four things carry most of the weight: the task, the context, the constraints, and the format you want back.
- Reading the critique and rewriting is where the learning happens, so run it more than once.
This runs a real language model, so results vary a little between runs. The critique is also written by a model, so treat it as a second opinion. Your prompts are sent to the model to produce a reply and are not kept after this page.