ChatGPT System Prompt — How to Set, Edit, and Customize It

The ChatGPT system prompt is the most important sentence you will ever write for an AI app. It controls the model's role, rules, tone, and behavior — for every response, in every conversation, forever. Get it right and the model behaves exactly the way you want. Get it wrong and you spend weeks chasing edge-case bugs in user-prompt logic that should have been fixed in two lines of system prompt.

This guide covers all three places you can set a ChatGPT system prompt, the structure of an effective one, and how to generate one from a free tool in two minutes instead of writing it from scratch.

Where you can set a ChatGPT system prompt

Where	How	Length	Use case
OpenAI API	role: "system" message	No hard limit	Custom apps, chatbots, agents
ChatGPT app	Settings → Customize ChatGPT	1500 chars × 2 fields	Personalize your own ChatGPT
Custom GPT	GPT builder → Instructions	8000 chars	Shareable assistants on the GPT Store
Projects	Project settings → Instructions	1500 chars	Per-project context
ChatGPT Enterprise	Admin console	No hard limit	Org-wide defaults

The API gives you the most control. The custom GPT instructions field is the second-best option for non-developers because it accepts a long structured prompt and can be reused by anyone you share the GPT with.

The 5-component anatomy of a good system prompt

Every effective system prompt has the same five parts. Skip one and the model fills in the gap with whatever it learned in training, which is rarely what you want.

1. Identity — Who the assistant is. "You are a senior customer support agent for Acme SaaS."
2. Capabilities — What it can do. "You answer billing questions, explain features, troubleshoot common errors, and escalate to humans when stuck."
3. Rules — What it must always do. "Always confirm the user's plan tier before discussing pricing. Always end with a yes/no question."
4. Constraints — What it must never do. "Never promise refunds. Never mention competitor product names. Never invent feature roadmap items."
5. Output format — How to respond. "Reply in 2-4 short paragraphs. Use bullet points for steps. End with a single clarifying question."

The free system prompt generator assembles all five components automatically. You pick a use case (chatbot, coder, writer, tutor, sales, legal, health, analyst, or custom), toggle the rules you want, and get a ready-to-paste prompt.

How to set a system prompt in the OpenAI API

In the API, the system prompt is the first message in the messages array, with role "system":

{
  "model": "gpt-4o",
  "messages": [
    {"role": "system", "content": "You are a senior customer support agent for Acme. Always confirm the user's plan before discussing pricing. Never promise refunds — escalate refund requests to a human."},
    {"role": "user", "content": "I want a refund for last month."}
  ]
}

The system prompt sticks for the entire conversation. Every user message is interpreted in light of it. You can update it between conversations but not mid-conversation (technically you can, but the model treats it as part of the user turn).

How to set custom instructions in the ChatGPT app

For personal use of ChatGPT.com (not the API), the equivalent is Custom Instructions:

1. Click your name in the bottom-left of ChatGPT
2. Select Customize ChatGPT
3. Fill in the two fields: "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?"
4. Save

These two fields combine into the system prompt for every conversation you start. Each field has a 1500-character limit, so keep it focused. The generator above produces concise prompts that fit comfortably.

How to set instructions in a custom GPT

Custom GPTs are the most powerful no-code option. Their "Instructions" field accepts up to 8000 characters and supports the full 5-component structure:

1. Go to chat.openai.com/gpts/editor
2. Click Configure
3. Paste your system prompt into the Instructions field
4. Add a name, description, and starter prompts
5. Save and test

Custom GPTs run on top of GPT-4o and inherit all of its capabilities. The system prompt you write here defines the entire personality and behavior of the GPT — users who chat with it never see the instructions.

What goes wrong when you write your own

The most common system prompt mistakes after reviewing hundreds of community prompts:

• Vague rules. "Be helpful" does nothing. "Always provide step-by-step numbered instructions" works.
• Contradictory instructions. "Be concise" + "give comprehensive explanations" confuses the model.
• Too many rules. Past 15-20 rules, the model starts forgetting some. Pick the 8-12 that matter most.
• No constraints. Without explicit "never do X" rules, the model will eventually do X for some user.
• Missing identity. "You are an AI assistant" is the default — useless. Be specific: "You are a senior tax accountant assistant for US small businesses."
• No output format. Without one, the model picks its own and varies between responses.

Generator output example

Picking the "Customer Chatbot" use case, adding the assistant name "Ada", company "TechStart", and toggling "stay on topic," "admit unknowns," "friendly," and "ask clarifying questions" produces:

You are Ada, a customer support assistant for TechStart.

You answer questions about the company, its products, pricing, and policies. You help users troubleshoot issues and guide them to solutions.

Rules:
- Stay focused on your area of expertise. If asked about unrelated topics, politely redirect the conversation.
- If you are unsure about something or don't have enough information, say so clearly. Never make up information.
- Use a warm, approachable tone. Be encouraging and supportive.
- When a request is ambiguous, ask a clarifying question before proceeding rather than guessing.

Always introduce yourself as Ada from TechStart in the first message of a conversation.

This is a complete, production-ready prompt that took 30 seconds to generate. You can paste it directly into the OpenAI API, a custom GPT, or any other LLM API.

Related tools and posts

• System Prompt Generator — Build a system prompt in 2 minutes
• Prompt Builder — Build user-level prompts for one-off tasks
• Token Counter — Check how many tokens your prompt uses
• AIPRM Alternative — A free prompt builder without the Chrome extension