System Prompt Generator for ChatGPT & Claude

Set the behavior you want across a whole conversation. Choose a starting example, edit the details and copy your system prompt as Markdown, XML tags or plain text.

FreeNo signupRuns in your browser

Anatomy of a system prompt

A system prompt sets standing behavior for a whole conversation. Give each instruction a clear home so you can edit one part without rewriting everything. The table uses a fictional support assistant as an example; replace its product details with information that is true for your situation.

Section	What goes in it	Example
Role	The job the assistant should perform.	Customer support agent for a project management app.
Context	Stable product details and the information available.	Use the supplied help articles and account details.
Audience	Who reads the answers and what they need.	Busy agency owners who need a clear next step.
Tasks	The recurring work, one action per line.	Give numbered troubleshooting steps.
Rules	Boundaries and behavior when information is missing.	Ask one clarifying question when the request is ambiguous.
Tone	The voice to maintain across replies.	Friendly and plain.
Output format	The expected shape of each answer.	Direct answer, steps if needed, then one next step.

Specific instructions beat general ones. “Give numbered troubleshooting steps” tells the assistant what to produce; “be excellent at support” gives you little to inspect. Prefer a positive action such as “state which detail is missing” over several overlapping prohibitions about uncertainty.

System prompt versus user prompt

Keep reusable behavior in the system prompt and the current assignment in the user prompt. A writing editor's standing instruction might ask it to preserve meaning and explain substantial changes. The user message should contain today's draft, its purpose and any exception that applies only to that draft.

The AI prompt generator helps you write that individual request. If you need to choose where to run it, use the model picker. Changing models is a reason to retest your instructions, so keep a small set of example requests with the prompt.

How to use the generator

Choose the nearest preset. Each one is an editable example, so check every product detail and task before copying it.
Write a precise role and add only context that should apply throughout the conversation. Put separate tasks and extra rules on separate lines.
Select rules that solve a real need. Keep the tone and output format consistent with those rules.
Pick a layout and copy the result. Empty sections are omitted, and changing the layout preserves your field values.

Markdown headings are widely used and work across models. XML-style tags are a format Anthropic recommends for structuring Claude prompts. Choose the version you find easiest to maintain, and use the practical prompt engineering guide for a broader approach to organizing instructions.

You can adapt the result for custom instructions or a project instruction field. Chat apps have character limits, so check the count against the destination field before pasting. The layout controls the structure of your instructions; the output format section controls the answers you request.

How to test and tighten a system prompt

Try the three requests most likely to break it. For support, that could be an unclear error report, a question with missing policy details and an unrelated request. Write down the behavior you expected before running each test, then compare the answer with that expectation.

Add a rule only for a failure you actually saw. If an answer invented a refund policy, make the source restriction explicit and rerun the same request. Avoid adding a page of restrictions after one disappointing result. Remove repeated rules, turn vague advice into concrete actions and check that a new rule does not conflict with an older one.

Frequently asked questions

What should I put in a system prompt?

Put the role, recurring context, rules and response format in the system prompt. Keep details that change with each request in the user prompt. Specific instructions give the model clearer direction than broad phrases such as be helpful.

Can I use this as a custom instructions generator?

Yes, you can adapt the output for ChatGPT custom instructions, a Claude project instruction or a Gemini Gem instruction. Check the character count against the instruction field in your app. Keep the most useful standing rules and remove context that belongs to a single request.

Should I use Markdown or XML tags for a system prompt?

Choose the layout that makes your instructions easiest to inspect and maintain. Markdown headings are widely used across models, and Anthropic recommends XML-style tags for structuring Claude prompts. The three layouts here carry the same section content.

What is the difference between a system prompt and a user prompt?

A system prompt holds standing instructions that apply to every message in a conversation: the role, the rules and the response format. A user prompt is the individual request. Put anything that changes from one request to the next in the user prompt and keep the system prompt stable.

How do I make a system prompt more reliable?

Test the three requests most likely to expose a weakness. Add a specific instruction for a failure you actually observed, then rerun those requests. Prefer a clear positive direction over a long list of prohibitions.

Does this tool use AI to write my instructions?

No, this free tool assembles your fields and selected rules in your browser. It does not run a model or test the instructions for you. Copy the result into your chosen app and check its behavior with real requests.