CraftPrompts

System prompt generator

For custom GPTs, Claude Projects, support bots and agents: describe the job and the rules once, and get standing instructions written in second person with a safe fallback.

Start from

Your prompt
roletaskcontextformattonerulesexample
Fill in a few fields and your prompt is set here, line by line.
0 words · 0 characters

Prompt score

Underspecified

0 words

  • Clear task verb missingStart with an explicit action: “Write…”, “Summarize…”, “Compare…”.
  • Background or context missingSay who it is for and why you need it; models fill gaps with generic guesses.
  • Role or perspective missingGive a perspective: “You are a senior UX writer…”. It sets vocabulary and depth.
  • Output format stated missingName the shape of the answer: table, bullets, JSON, 3 paragraphs.
  • Length or scope limit missingCap the length: “under 150 words”, “5 bullets”.
  • Constraints or rules missingAdd rules: what to avoid, what must be included, reading level.
  • Example or reference missingShow one example of the output you want. Few-shot beats description.
  • Specific, not vague missingReplace vague words (good, nice, interesting) with concrete criteria.

Let AI write the system prompt

Sends your filled-in fields to the AI helper, which writes a complete prompt and marks gaps as [placeholders].

The parts of a solid system prompt

  1. Identity and job. “You are the returns assistant for Acme. Your job is…”
  2. Audience. Who is on the other side, and what they usually know.
  3. Knowledge. Facts, policies or product data the assistant can rely on.
  4. Response style. Tone, format, length. Short is usually better in chat.
  5. Rules. What it must never do, and when to hand over to a human.
  6. Fallback. What to do when a request is out of scope or unclear.

Test it with awkward questions before you ship: off-topic requests, angry users, requests for things the rules forbid. Adjust the rules until the answers are what you would say yourself.

Questions

What is a system prompt?

A system prompt is the standing instruction a chat model receives before any user message. It defines who the assistant is, what it does, how it answers and what it refuses. Custom GPT instructions, Claude Projects instructions and the “system” message in the API are all system prompts.

How long should a system prompt be?

As long as it needs to be to remove ambiguity, typically 150 to 600 words for a focused assistant. Long prompts are fine if they are organised; a rambling one produces inconsistent behaviour.

Can a system prompt stop jailbreaks?

It helps but it is not a security boundary. Clear rules and a fallback (“if unsure, ask”) reduce off-topic answers, but never put secrets in a system prompt and validate anything important in your own code.

Why add “if unsure, ask a clarifying question”?

Models tend to answer confidently even when a request is ambiguous. An explicit fallback gives them permission to ask instead of guessing, which is usually what a user wants.

More tools in the studio