# System Prompt Examples: 7 to Copy for Claude and ChatGPT > System prompt examples you can copy for Claude and ChatGPT: writing, support, coding, research and more, plus the five parts every good system prompt has. Canonical: https://nexin.si/blog/system-prompt-examples [Guides](https://nexin.si/blog/category/guides) Nexin · November 3, 2026 · 6 min read A system prompt is the standing instruction a model reads before every message in a chat. It says who the model works for, what it should do, how it should answer and what it must never do. Good system prompt examples share five parts: a role, the context, the task rules, the output format and the limits. Below are seven you can copy and adapt for Claude or ChatGPT, then a short guide to writing your own. ## What is a system prompt? It’s the instruction that sits above the conversation. Your normal messages ask for one thing; the system prompt sets the rules for all of them. Change it once and every answer in that chat changes with it. Where you set it depends on the tool: | Where | What it’s called | |---|---| | Claude API | The `system` parameter | | Claude.ai Projects | Project instructions | | Claude Code | `CLAUDE.md`, read at the start of each session (not technically a system prompt, but it does the same job) | | ChatGPT | Custom instructions, or a custom GPT’s instructions | | OpenAI API | A `system` or developer message | Anthropic publishes the system prompts it uses for the Claude apps in its release notes. They’re worth reading: long, and mostly about behavior, not personality. ## What makes a good system prompt? Most weak system prompts are one line: “You are a helpful marketing expert.” The model already knows how to be helpful. What it doesn’t know is your business, your reader and your rules. A system prompt that works answers five questions: 1. **Role:** who is the model working for, and as what? 2. **Context:** what does it need to know that it can’t guess (product, audience, tone)? 3. **Rules:** what does it always do, step by step? 4. **Format:** what should the answer look like? 5. **Limits:** what must it never do, and what should it do when it doesn’t know? Write in plain sentences, put the most important rule first, and say why a rule exists when it isn’t obvious. Models follow a reason better than a bare “never”. ## System prompt examples you can copy Replace everything in [brackets]. Each example is a starting point; the context section is where yours will differ most. ### 1. Writing assistant in your voice ``` You help [name] write posts, emails and pages for [business], which sells [offer] to [audience]. Voice: plain and direct, short sentences, no hype words, no exclamation marks. Write the way [name] talks to a customer, not like an ad. Before writing, ask what the piece is for and who reads it if I haven't said. Use only facts I give you. If a number, quote or result is missing, put it in [brackets] instead of inventing it. Give one draft, then a list of anything you had to assume. ``` ### 2. Customer support replies ``` You draft replies to customer emails for [product]. A person reviews every reply before it is sent. Facts you may use: [pricing, refund policy, shipping times, known issues]. If the answer isn't in these facts, say so in a note to me and don't guess. Each reply: thank them once, answer the question in the first two sentences, give the next step, sign as [name]. Under 120 words. Match the customer's language. Never promise refunds, dates or features that aren't in the facts above. ``` ### 3. Coding helper ``` You are pairing on a [language/framework] project. The code style is [style notes]. Read the code I paste before suggesting changes. Prefer the smallest change that fixes the problem. Explain the cause in one or two sentences, then show the change as a diff. If a change needs a new dependency, say so and ask first. If you're not sure the code runs, say what you'd test. ``` For Claude Code, the same rules belong in a `CLAUDE.md`. The [CLAUDE.md Starter](https://nexin.si/library/claude-md-starter) template has a complete one, and our post on [CLAUDE.md](https://nexin.si/blog/claude-md) explains each part. ### 4. Research summarizer ``` You summarize documents I paste for [purpose, e.g. a weekly briefing for the team]. Start with the three points that matter most for [purpose], each in one sentence. Then the details, grouped by topic. Quote the source line for any number. Mark anything uncertain or contradictory. Don't add facts that aren't in the document. ``` ### 5. Social media content planner ``` You plan social posts for [brand] on [platforms]. Audience: [who]. Goal: [followers, leads, sales]. Content pillars: [3 to 5 topics]. Each post: a hook in the first line, one idea, a clear ending. Mix formats across the week. Output a table: day, platform, pillar, hook, format. Then write the posts I pick. ``` Our guide to [content pillars](https://nexin.si/blog/content-pillars-for-social-media) helps fill in that part. ### 6. Editor that only edits ``` You edit my drafts. You do not rewrite them in your own style. Fix grammar, cut filler, and flag sentences that are unclear. Keep my words wherever they work. Show the edited text, then a short list of what you changed and why. Never add new claims, examples or facts. ``` ### 7. Tutor that makes you think ``` You tutor me in [subject] at [level]. Don't give the answer first. Ask one question at a time that moves me toward it. If I'm stuck twice, give a hint; if I'm stuck three times, explain and then give me a similar problem to try. Keep explanations short and check that I understood before moving on. ``` ## How do you write your own system prompt? Start from your best conversation. Look back at a chat where the model did exactly what you wanted, and list what you had to tell it along the way. Those corrections are your system prompt. Then test it on five real requests, including one awkward one. When an answer goes wrong, add one rule that would have prevented it. Keep the prompt as short as the rules allow; a page of conflicting instructions works worse than ten clear lines. If you don’t want to start from a blank page, the [AI Prompt Generator](https://nexin.si/library/ai-prompt-generator) prompt turns a rough request into a full prompt with role, context, steps, format and checks. ## System prompt or skill? A system prompt applies to every message in a chat. A skill is a set of instructions Claude loads only when a request matches it, so you can have many without crowding the conversation. Rules for how you always want answers go in the system prompt; a method for one kind of task, like writing an email or a carousel, works better as a skill. More in [what Claude skills are](https://nexin.si/blog/what-are-claude-skills) and [Claude Projects vs skills](https://nexin.si/blog/claude-projects-vs-skills). ## FAQ ### What is the difference between a system prompt and a user prompt? The system prompt sets standing rules for the whole conversation. A user prompt is one message asking for one thing. The model reads both, with the system prompt as the frame. ### How long should a system prompt be? As long as your rules need and no longer. Many good ones are 100 to 300 words. Add a rule when you see a mistake, and remove rules that no longer change anything. ### Can I see ChatGPT’s or Claude’s system prompt? Anthropic publishes the system prompts for the Claude apps in its release notes. OpenAI doesn’t publish ChatGPT’s in the same way. ### Do system prompts work in the free plans? Custom instructions in ChatGPT and Projects in Claude.ai have plan limits that change over time, so check each app’s current plan page. Through the APIs, the system prompt is always available. --- **Skip the blank page.** The Nexin library has skills, prompts and templates for Claude, like the [Email Writer](https://nexin.si/library/email-writer) and [Humanize AI Text](https://nexin.si/library/humanize-ai-text) skills and the [CLAUDE.md Starter](https://nexin.si/library/claude-md-starter), plus kits for Claude Code. Guides like [Install a Skill in Claude](https://nexin.si/library/install-a-skill-in-claude) are free. [Get Nexin, $27 once](https://nexin.si/checkout?item=nexin:once), with a 7-day money-back guarantee. **About this article.** Written by the team that builds Nexin, a content system of tools and prompts for Claude and ChatGPT. Spotted something wrong or out of date? Write to [support@nexin.si](mailto:support@nexin.si). ## Turn one idea into every channel. [See pricing](https://nexin.si/#pricing) More in Guides [GUIDES](https://nexin.si/blog/category/guides) Oct 26, 2026 · 6 min ## [How to Write a PRD with AI: A Step-by-Step Guide](https://nexin.si/blog/how-to-write-a-prd-with-ai) How to write a PRD with AI in Claude or ChatGPT: what to give the model, the sections to ask for, a prompt that works and how to check the draft. [GUIDES](https://nexin.si/blog/category/guides) Oct 26, 2026 · 6 min ## [Business Plan with ChatGPT: How to Write One That Holds Up](https://nexin.si/blog/business-plan-with-chatgpt) How to write a business plan with ChatGPT or Claude: what to give it, the sections to ask for, a prompt you can paste, and the numbers to check yourself. [GUIDES](https://nexin.si/blog/category/guides) Oct 22, 2026 · 6 min ## [Humanize AI Text: How to Make AI Writing Sound Human](https://nexin.si/blog/humanize-ai-text) How to humanize AI text by hand or with Claude: the patterns that make writing sound machine-made, what to cut, what to add, and a before-and-after example.