AI persona prompt generator
Assemble a reusable system prompt that defines an AI persona from a few short fields.
AI persona prompt generator
Writing a good system prompt from scratch is fiddly: it is easy to forget a section, phrase the tone inconsistently, or bury the one rule that matters under a wall of prose. This tool assembles one for you. Pick a role, a tone, an audience, an output format, a response length and a reply language, and it returns a clean, sectioned system prompt ready to paste into any chatbot, custom GPT, API call or agent framework.
Choose from thirteen roles — senior developer, copywriter, editor, teacher, analyst, product manager, translator, support agent — or select "Custom" and describe the role yourself. Type a domain or topic into the second input (the first line, e.g. "react and typescript" or "customer refunds for a SaaS product") and the generator writes it into the persona's stated focus. Further lines in that box become extra rules, carried into the prompt verbatim under an "Additional rules" section — useful for a house style guide, a list of banned phrases, or a formatting rule specific to your use case.
The output format option covers prose, bullet points, Markdown and strict JSON-only replies, and the response-length option keeps the model from padding answers when you want brevity or from being terse when you need a thorough one. Turn on the boundaries clause to add an instruction to decline or redirect out-of-scope requests instead of guessing, and turn on the few-shot placeholder to drop in a ready-made "## Examples" block you can fill with your own sample exchanges. A structure option emits the same sections as Markdown headings, XML tags or plain titles, and further switches make the persona admit uncertainty, ask a clarifying question, reason step by step, avoid emoji, or end on an {{input}} placeholder.
Everything happens instantly in your browser as you change options — no API call is involved, and nothing you type is uploaded or logged anywhere. Copy the result, download it as a .txt file, or send it straight into another tool to keep refining it.