AI Prompt Generator & Optimizer
Paste a prompt and see what it's missing in seven checks, then let AI rewrite it and show you every change. Or describe what you need and generate a system prompt, task prompt or few-shot template. Works for ChatGPT, Claude and Gemini.
AI Summary
Free AI prompt optimizer and generator. A local prompt check lists which of seven parts a prompt is missing (role, task, format, length, audience, constraints, examples) with a fix for each. An AI rewrite improves the prompt in four styles and shows a word-level diff. Generate mode writes a prompt from a description.
Input: A prompt to improve, or a plain-English description of the prompt you need (up to 5,000 characters)
Output: An improved or generated prompt, a word-level diff, a checklist, and an estimate of tokens and cost per 1,000 runs
Privacy: The prompt check runs in your browser. The AI rewrite sends your text through Dervity's server. We don't save your prompts.
- Prompt check for role, task, format, length, audience, constraints and examples
- Rewrite styles: Clearer, More detailed, Shorter, Structured
- Changes view with a word-level diff of the rewrite
- Generate system prompts, task prompts, step-by-step and few-shot templates
- Token estimate and input cost per 1,000 runs on current models
⌘ Enter
Removes vague wording and adds the context the model would otherwise guess.
Your prompt
The prompt check below runs as you type, without AI.0/5,000
Improved prompt
Your improved prompt shows up here
Pick a style and press Improve with AI. Then open Changes to see every word that moved.
The AI rewrite runs through our server. We don't save your prompts.
Prompt check
No AI, runs in your browser
Seven things a good prompt spells out
Role, task, format, length, audience, constraints and examples. Paste a prompt and see which ones are there.
Tokens and cost
- Your prompt
- Nothing yet
- Cost per 1,000 runs
- Rewrite
- Nothing yet
- Cost per 1,000 runs
Input only, from list prices. Replies are billed on top at $10 per million output tokens. Token counts are estimates (about 4 characters each).
Was this tool helpful?
What makes a prompt better?
A model fills every gap in your prompt with an average guess. The same request with its gaps filled in, run through the same check the tool uses.
Write me a blog post about productivity
- Role or context (missing)
- Clear task verb (in place)
- Output format (missing)
- Length (missing)
- Audience or tone (missing)
- Constraints (missing)
- Examples (missing)
You are an editor for a blog read by engineering managers at software companies. Write a blog post about protecting focus time on a remote team. Audience: managers of 5 to 15 engineers. Tone: practical and friendly, no hype. Length: about 900 words. Format: Markdown with a short intro, 4 H2 sections and a two-sentence ending. Rules: - Every tip must be something a manager can do this week. - Avoid generic advice like "take breaks" or "use a to-do list". Example of the level of detail I want: "Block Tuesday and Thursday mornings as no-meeting time and move standups to 11:30."
- Role or context (in place)
- Clear task verb (in place)
- Output format (in place)
- Length (in place)
- Audience or tone (in place)
- Constraints (in place)
- Examples (in place)
How to optimize a prompt
Paste your prompt
The check lists which of the seven parts are missing and how to add each one. You can fix them by hand and stop there.
Pick a style and improve
Clearer, More detailed, Shorter or Structured. The rewrite streams in, and Changes shows each word added or cut.
Check the cost and copy
Compare tokens before and after and the input cost per 1,000 runs, then copy it or press Edit to keep refining.
Which prompt type should you generate?
Generate mode writes a prompt from a description. The type decides what the prompt includes and where you'd paste it.
| Use it for | What you get | |
|---|---|---|
| System prompt | Custom instructions, a GPT, a Claude project, an API system message | Role, rules, output format and limits |
| Task prompt | One piece of work in a chat | Context, instructions, format and what a good answer looks like |
| Step by step | Math, logic, planning and debugging | Instructions to reason through steps before answering |
| Few-shot | Classifying, extracting or rewriting in a fixed pattern | 2 or 3 example inputs and outputs, then your input |
Make the first draft sound less like AI
A good prompt gets you the right content. This free cheat sheet fixes the voice: the words and habits that make AI text read as machine-written, plus a style prompt you can add to any chat.
- 60 words AI overuses, with what to write instead
- 12 sentence patterns that give AI text away
- A copy-paste prompt for a less robotic first draft
- A 10-point check to run before you publish
Picked the prompt? Pick the model.
Input prices on current models run from $0.10 to $10.00 per million tokens, so the same prompt can cost 100 times more on one model than another.
Questions
How do I optimize an AI prompt?
Paste the prompt, read the prompt check, then press Improve with AI. The check shows which of seven parts are missing: role, task, format, length, audience, constraints and examples. The AI rewrite fills the gaps in the style you pick, and the Changes tab shows every word it added or cut, so you can keep only what you agree with.
What does the prompt check look for?
It looks for seven things a model can't guess: who it is or what it's working on, a task verb, an output format, a length, the audience or tone, rules to follow and an example. Each one is a pattern in your text, so the check runs in your browser without AI and tells you the one-line fix for anything missing.
How do I generate an AI prompt from a description?
Switch to Generate from a description, pick a prompt type and describe what you need in a sentence or two. System prompt gives you a role, rules and an output format. Task prompt is for a single job. Step by step asks the model to reason first. Few-shot adds 2 or 3 example inputs and outputs.
Which rewrite style should I pick?
Pick Clearer when the prompt is vague, More detailed when answers come back generic, Shorter when a long prompt repeats itself or costs too much per call, and Structured when the model mixes up instructions or you need a fixed output shape. You can run the same prompt through two styles and compare them in the Changes tab.
Do the improved prompts work with ChatGPT, Claude and Gemini?
Yes. The rewrite is plain text with no model-specific syntax, so it works in ChatGPT, Claude, Gemini, Copilot and through any API. Every major model gives steadier answers when the prompt states the format and length, and the Structured and More detailed styles add both. To pick the model itself, use the Model Selector.
What's the difference between a system prompt and a task prompt?
A system prompt sets the role, rules and output format for a whole conversation or app, and it's sent with every request. A task prompt asks for one piece of work, with the context and details for that job. Use Generate with the System prompt type for custom instructions or an API system message.
How much does a longer prompt cost to run?
Every token in the prompt is billed as input on every call. The Tokens and cost card estimates the tokens before and after the rewrite and the input cost per 1,000 runs on the model you pick, from list prices. A 500-token system prompt on a model at $2 per million input tokens costs $1 per 1,000 calls before the reply.
Is the prompt optimizer free, and are my prompts stored?
It's free with no signup. The prompt check runs in your browser. The AI rewrite sends your text through our server to a writing model we pick for quality and cost. It allows about 10 rewrites a minute per visitor. We don't save the prompts you send.
Related tools
More free tools
Everything from model picks to CSV converters, with no account needed.