System Prompt Generator

What Is the System Prompt Generator?

The System Prompt Generator is a form that writes the opening instruction you give a language model before your actual request. Instead of staring at an empty chat box, you fill in who the model should be, what it should do, who it is talking to, how it should answer and what it must never do. The generator turns those answers into a clean, sectioned prompt you can paste into ChatGPT, Claude, Gemini, a local model or an API system message.

It is a builder, not a model. Every word in the output comes from your inputs, and only the sections you filled appear. The whole thing runs in your browser, so your draft is never transmitted anywhere, and it is auto-saved locally so you can close the tab and pick up later.

Why You Would Use It

A system prompt is the cheapest way to make model output consistent. Without one, the same question can return a table today and a paragraph tomorrow, with a different tone every time. With one, the role, format and rules are fixed before the first user message, which matters when the prompt sits behind a product, a support bot or a content pipeline.

This tool suits developers wiring an assistant into an application, support teams standardising answers, marketers who need the same voice across a hundred generations, teachers preparing a tutoring persona, and non-technical users who want the structure without memorising a prompt recipe. The eleven presets cover common jobs, and each one pre-fills every field so you start from a working prompt rather than a blank page.

How This Tool Works

The left column is the form. It opens with a preset grid: Blank plus eleven starting points (Coding assistant, Writing coach, Data analyst, Marketing copywriter, Customer support agent, SEO specialist, Translator, Academic researcher, Sales email writer, Resume and interview coach, Productivity coach). Clicking a preset fills every field with values chosen for that job. Clicking Blank clears the form while keeping the language on English.

Role and Task description are the two required fields; everything else is optional. Audience tunes who the response is for. Expertise level, Tone and Output format are chip groups where one choice is active at a time, Language is a dropdown of ten languages, Constraints is a textarea where each line becomes one bullet, and Behaviour is a multi-select of seven rules such as asking a clarifying question when a request is ambiguous, saying it does not know instead of guessing, reasoning step by step, showing trade-offs, listing assumptions, citing sources, and flagging security issues.

The right column renders the prompt live as you type. Sections are emitted in a fixed order: Role (You are ...), Task, Audience (Address ...), Expertise level, Tone, Constraints, Output format, Language and Behaviour. Empty sections are skipped completely, so the prompt stays as short as the inputs justify. Leading list markers on constraint lines are stripped and empty lines dropped, which keeps the bullets tidy.

Under the output you get live counters for characters, words, approximate tokens and the number of sections, plus a constraint count beside the Constraints label. A badge in the header reads Incomplete until both Role and Task are filled, then flips to Ready to copy; the two required inputs are highlighted as invalid while they are empty but output exists. Your draft is written to local storage on every change and restored on the next visit.

Key Features

• Eleven presets plus a blank start, twelve starting points in total
• Live sectioned output that omits every empty field
• Six tones, five output formats, five expertise levels and ten languages
• One-per-line constraints rendered as clean bullets
• Seven behaviour rules you can toggle on or off independently
• Character, word, token and section counters, plus a constraint count
• Copy prompt, Copy as JSON and Download as .txt
• Ctrl/Cmd+Enter shortcut that copies the prompt from anywhere on the page
• Load example, Paste into task and Reset controls
• Draft auto-save in the browser with restore on your next visit

How to Use the Tool

1. Pick a preset card that matches the job, or click Blank to describe everything yourself. Press Load example if you want to see a complete prompt first.
2. Fill Role and Task description; both are required and the badge stays Incomplete until they exist. Add Audience if you know who reads the answer.
3. Tap one chip each for Expertise level, Tone and Output format, then choose the Language the model should reply in.
4. Type Constraints one per line. Every line becomes a bullet in the prompt, and the counter beside the label shows how many you have.
5. Switch on any Behaviour rules you want, then read the prompt in the right-hand panel.
6. Press Copy prompt, Copy as JSON or Download .txt, or simply hit Ctrl+Enter (Cmd+Enter on macOS). Use Paste into task to pull a brief straight from your clipboard, and Reset to start over.

Limits and Rules

Only Role and Task description are required. Everything else is optional and simply disappears from the output while empty, which is why a prompt can change length as you fill it in. Language defaults to English and Reset returns it to English.

Constraints are taken line by line: leading dashes, bullets and asterisks are removed and blank lines ignored, so write plain rules and let the tool add the bullet. Behaviour entries are copied using their labels from the configuration, so you get the same wording every time.

There is no size limit, no rate limit and no captcha, because the generator makes no network requests at all. The trade-off is that your draft lives only in local storage on this device, and clearing site data wipes it. The token figure is an estimate (characters divided by four), not a tokenizer result, and the output is plain text with simple headings, so it cannot enforce a schema for you.

When to Use It and When Not to

Use it when you need the same behaviour from a model many times, when a team should share one agreed instruction, when you are moving a prompt from a chat window into an API system message, or when you want to see how much a prompt costs in tokens before you commit to it.

Do not use it as a place to put large documents or few-shot examples; it builds the instruction, not the data. It also does not test the prompt, run the model or grade the answers. If you need per-request logic such as switching persona by user type, that belongs in your application code rather than in a static system prompt.

Frequently Asked Questions