Turn a rough idea into a prompt any AI understands

How the AI prompt generator builds your prompt

Most weak AI answers come from prompts that leave the model guessing: no role, no audience, no format. The generator fixes that by wrapping your idea in the parts a good prompt needs. First it reads your text and detects the task type (coding, email, SEO, social media, career, data, marketing, teaching, research, business, creative writing or general writing) from the words you use. Each task type brings its own expert role and requirements. A request about a cover letter, for example, gets a recruiter persona and rules like "use action verbs and measurable results" and "never invent experience".

Then it applies the options you chose. Tone adds a single clear instruction such as "Use a persuasive tone." The answer format adds a precise output spec, for instance a Markdown table with column headers, valid JSON with descriptive keys, or an email with a subject line and a call to action. The answer language adds "Respond in Spanish" (or English, German, French, Portuguese, Italian or Japanese), which is handy when you write the idea in one language but need the result in another.

Simple, Advanced or Expert: which detail level to use

• Simple gives you a role line, your request and up to three key requirements. Use it for quick everyday questions where a long prompt would be overkill.
• Advanced (the default) splits the prompt into Role, Context, Task and Requirements, plus Output format when you pick one. The Context section contains a [bracketed] placeholder: replace it with background the AI cannot guess, such as who you are or what you already tried.
• Expert adds an Example slot for a sample of the style you like and a Quality check list. It also tells the AI to ask up to three clarifying questions if critical information is missing. Choose it for important work: proposals, long articles, code, analysis.

If you are unsure, start with Advanced, run it, and switch to Expert only when the answer still feels generic.

One prompt, eleven target AIs, different formatting

The same content is formatted differently depending on the assistant you select, because each model responds best to its own conventions:

• Claude: every part goes inside XML tags such as <role>, <task> and <requirements>, the structure Anthropic recommends.
• ChatGPT, Microsoft Copilot and Grok: Markdown headings (## Role, ## Task), which these chat interfaces read cleanly.
• Gemini: plain labeled sections in capitals (ROLE:, TASK:) without Markdown symbols.
• DeepSeek, Qwen, Llama and Mistral: Markdown plus a line asking the model to think through the problem step by step and present only the polished final answer.
• Perplexity: an extra line asking it to search the web, prefer sources from the last 12 months and cite them inline.
• Any AI / Universal: neutral Markdown that works almost everywhere.

For deeper advice on a single assistant, use the dedicated pages such as the Claude prompt generator or the ChatGPT prompt generator.

Voice input, history and the Open in buttons

The microphone button lets you dictate your idea instead of typing; it works in browsers that support speech recognition and shows a message when yours does not. Your last 20 prompts are kept in a history list stored only in this browser, so you can click an old idea to restore both the request and the prompt it produced. Clearing your browser data removes it.

Under the result, the Open in links launch ChatGPT, Claude or Perplexity with the prompt already passed in the link, so you skip the copy-paste step. Very long prompts are shortened in the link, so for those use Copy. Once you have a draft you like, run it through the prompt checker to see which ingredients are still missing, or check its size with the token counter. When AI mode is available on this site, an extra Enhance with AI button rewrites the local prompt with a language model while keeping every detail you gave.