What it’s for
- Run a generated/optimized prompt and see what it produces
- Iterate on wording, tone, constraints, and output format
- Save the best version back to My Prompts
What it’s not for
Prompt Assistant isn’t intended for completely unrelated/general chat. If you want help, share the prompt you’re testing (or your goal) and the output format you want.Choose a model
Select the assistant model you want to run the prompt with. If you’re unsure, start with the default and switch models only when you want a different style or speed.How it works
1
Start with a prompt
Paste a prompt, or insert one from My Prompts.Common sources:
- Prompt Generator output
- Prompt Optimizer output (via Run)
- A saved prompt from Prompt Library
2
Pick an assistant model
Choose which model should run the prompt (default model is configured in the app).
3
Run + iterate
Send the prompt and evaluate the response. Iterate with follow-ups like:
- “Keep the same structure, but make it shorter”
- “Return output as JSON with keys …”
- “Ask clarifying questions first”
- “Add 3 examples and 3 edge cases”
4
Save to My Prompts
Use Save to store the exact version you want in My Prompts so you can reuse it later.
Core concepts
Generator vs Optimizer (when to use which)
Generator vs Optimizer (when to use which)
- Prompt Generator: start from a rough idea and generate a first draft prompt, then refine it in chat.
- Prompt Optimizer: start from an existing prompt and upgrade it into a clearer, better-structured version (with history).
Assistant model
Assistant model
The model selector controls which assistant runs your prompt.
Insert prompts from My Prompts
Insert prompts from My Prompts
Use My Prompts in the composer to insert a saved prompt and continue iterating.
Pin important work
Pin important work
Use Pin to keep key prompts and conversations easy to find.
From Optimizer to Assistant
From Optimizer to Assistant
Using Run in Prompt Optimizer opens a new Assistant chat seeded with the optimized prompt.
Tips for better chats
- Tell the assistant what success looks like (format, length, constraints).
- Include examples of “good” vs “bad” outputs if you have them.
- If you want the assistant to ask questions first, say so explicitly.