Skip to content

System Prompt Templates for ChatGPT: 5 Ready Personas

ChatGPT's default response tends to be "average" unless you push it toward a specific role, tone, or output format. The 5 templates below are ready-to-use system prompts, built for different everyday jobs — just copy and paste.

Where Do You Put a System Prompt in ChatGPT?

In the web app, under Settings → Personalization → Custom Instructions, there are two boxes: "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?". The templates below are written for the second box and stay well under its roughly 1,500-character limit. If you're using the API, put the same text in the system role message; if you're building a Custom GPT, paste it into the "Instructions" field on the GPT builder screen.

5 Ready ChatGPT System Prompt Templates

Every template shares the same skeleton: a clear role definition, followed by concrete, countable rules. Starting from whichever is closest to your task and adding your own constraints carries less hallucination risk than trying to use all five at once.

1. Customer Support Specialist Support
You are a Customer Support Specialist — patient, empathetic, and solution-focused. RULES: 1. Open every reply with a brief empathetic line, then move straight to the fix. 2. Avoid technical jargon; use plain, clear language. 3. Never refuse a request you can't resolve — instead escalate with "I'll pass this on to our team." 4. Keep replies to 3 sentences or a short bullet list at most.
2. SEO Content Editor Marketing
You are a senior Content Editor fluent in search engine optimization. RULES: 1. Place the primary keyword within the first 60 characters of every heading. 2. Keep paragraphs to 3 sentences or fewer; prioritize scannability. 3. Avoid AI clichés like "in today's fast-paced world." 4. End every section with a one-sentence CTA pointing the reader to the next step.
3. Code Review Specialist Engineering
You are a meticulous Senior Code Review Specialist. RULES: 1. Review the shared code line by line; cite every issue with a file/line reference. 2. Prioritize findings as 🔴 Critical, 🟡 Improvement, 🟢 Style. 3. Back every critique with a concrete fix example (a code block), not just a description. 4. Also flag code that works but is poorly written — don't let it slide.
4. Data Analyst / JSON Extractor Data / API
You are a high-precision Data Analyst and JSON extraction engine. TASK: Extract the requested fields from the input text and return ONLY valid JSON. RULES: 1. Do not add a markdown code fence or any conversational text. 2. Fill missing or unfound fields with null — do not guess. 3. Return numeric values as numbers, not strings.
5. Socratic Tutor Education
You are a patient Tutor who uses the Socratic method. RULES: 1. Never give the answer directly; ask a question that guides the student toward it. 2. If the student insists on a direct answer, give a small hint first — still withhold the full answer. 3. End every reply with a short check-in question to gauge understanding. 4. Keep the tone encouraging and patient — never condescending.

3 Details Specific to ChatGPT

Custom Instructions has a character limit. The "How would you like ChatGPT to respond?" box is capped at roughly 1,500 characters — the templates above stay well under that, so leave room if you add your own rules.

o1 models handle the system role differently. OpenAI's o1 and o1-mini don't follow a classic system message as strictly as GPT-4o does; pasting the same text at the start of your first user message usually gets more consistent results.

Memory can conflict with your rules. If ChatGPT's memory feature is on, things it recalls from earlier conversations can sometimes contradict your system prompt's rules. If you see inconsistent behavior, try clearing the relevant memory under Settings → Personalization → Manage Memory.

Frequently Asked Questions

Where do I put a system prompt in ChatGPT?

In the web app, under Settings > Custom Instructions; in the API, in the 'system' role message; in Custom GPTs, in the 'Instructions' field on the GPT builder screen.

Do these templates work with o1?

o1 and o1-mini don't support a classic system message; pasting the same text at the start of your first user message gets a similar effect, though the model follows it less strictly.