Bible Network Crypto DeFi Onchain RWA AI Agent Stablecoin Chain SAFU CryptoTax DeFAI AGI Claude Me Claude Skill Claude Design Claude Cowork
Independent Media
Not affiliated with any project
Exploring the Frontier of AI Intelligence
claude-me.com
LATEST
2026 Claude Model Family Deep Dive: What's New, When to Switch, and What It Costs  ·  Claude API Production Deployment: Engineering Checklist from Prototype to Stable Launch  ·  Five Common Claude Mistakes Beginners Make (And How to Fix Them)  ·  Claude Enterprise vs Team: Which Plan Does Your Company Actually Need? Past This Scale You Must Upgrade  ·  Using Claude for Deep Research and Knowledge Synthesis: From Multi-Source Information to Opinionated Analysis Reports  ·  Mechanistic Interpretability: Why Anthropic is Dissecting Claude's 'Brain' — Frontier AI Explainability Research
Glossary · prompt-techniques

Prompt vs System Prompt

prompt-techniques 新手

30-Second Version · For the impatient
A "prompt" is the question or instruction you input in the conversation; a "System Prompt" is the "background rules" set before the conversation starts, determining Claude's role, behavioral guidelines, and constraints in this conversation. Analogy: a prompt is you telling Claude "do this thing"; a System Prompt is you telling it before the conversation "who you are and how you should behave." For regular claude.ai users, this difference is most visible in Claude Projects' Instructions settings.
Full Explanation +
01 · What is this?

Prompt and System Prompt are two different layers of input in AI conversation:

Prompt: everything you type in the conversation box — your questions, instructions, pasted document content. Prompts determine "what Claude should do this time." Each input is a new prompt affecting that response.

System Prompt: framework rules set before the conversation begins — telling Claude its role in this conversation, rules to follow, and action boundaries. System Prompt applies continuously throughout the conversation; regardless of how many rounds follow, rules set in the System Prompt remain active.

Concrete example: System Prompt: "You are a professional financial advisor. Only discuss financial-related questions. Use formal but not overly academic tone. Maximum 200 words per response." Prompt: "Please explain what compound interest is." Claude's behavior in this setup: formal but not academic tone, within 200 words, explaining compound interest. If you ask an unrelated question ("write me a poem"), Claude might say "this is outside my scope as a financial advisor" — because System Prompt set its role boundary.

Where claude.ai users encounter System Prompt most: Claude Projects' Instructions field is the System Prompt. What you write there automatically loads every time you open that Project's conversation.

02 · Why does it exist?

How do you write a good System Prompt? Are there universal structure recommendations?

Good System Prompts have several core elements:

Role and context (Who): tell Claude who it is and what context it works in. "You are a consultant with 10 years of marketing experience serving early-stage tech startups" is more precise than "you're a marketing consultant" and produces more targeted responses.

Behavioral rules (How): how Claude should act in this conversation — language (Chinese or English), tone (formal/casual), response length (brief/detailed), format preference (bullets/paragraphs).

Scope limits (What): what Claude should and shouldn't do. For a legal Q&A assistant: "Only discuss legal questions; medical and financial questions are outside scope; suggest users consult professionals for these."

Special knowledge or context (Context): background information Claude needs to know for this Project — your company, users, products, style guides.

Simple System Prompt template: "You are [role], serving [audience]. Your primary task is [core function]. Please respond in [language], [formal/casual] tone, [length requirements]. For [out-of-scope situations], please [handling approach]."

Most important principle: don't make System Prompt too long. Put truly important rules in; secondary details can be explained in-conversation. A streamlined 200-500 token System Prompt usually works better than a detailed 2,000-token one, at lower cost.

03 · How does it affect your decisions?

Are Claude Projects Instructions and the API's System Prompt the same thing?

Functionally the same, but different usage and scenarios:

Claude Projects Instructions: used in the claude.ai interface. Instructions written in Project settings automatically load every time you open a conversation in that Project. This is how regular claude.ai users use System Prompt functionality — no technical knowledge required.

API System Prompt: in API calls, passed via the system parameter. Developers precisely control System Prompt content in each API request — can dynamically modify, set different System Prompts for different users, combine with Prompt Caching to reduce costs.

Commonalities: effect on Claude is identical — System Prompt sets the framework before conversation begins, applies throughout, has higher priority than user prompts.

Selection recommendation: regular claude.ai users → use Claude Projects Instructions for usage context setup. API developers → use API system parameter to precisely control each request's System Prompt.

04 · What should you do?

If my prompt conflicts with the System Prompt, which takes priority? Is there a way to override System Prompt settings during conversation?

General rule: System Prompt has higher priority than prompts.

When your prompt conflicts with System Prompt (e.g., System Prompt says "only use Traditional Chinese" but you ask in English), Claude usually follows System Prompt rules and answers your English question in Traditional Chinese.

But not absolute: System Prompt sets "default behavior and constraints," not "final decision for all cases." Exceptions: if System Prompt rules are clearly unreasonable ("never admit to being an AI"), Claude still maintains honesty. If you explicitly say in your prompt "for this one please ignore the language constraint and answer in English," Claude usually honors your explicit request as a "conscious preference adjustment" not "accidentally violating rules."

API developer control: in API, you can dynamically modify System Prompt on each call, or add override instructions in the user message.

Advice for claude.ai users: if your Instructions settings make conversations inflexible (language restrictions you constantly work around), directly modify Instructions — rather than repeatedly saying "please ignore the previous rule" in prompts. The latter is inefficient; the former is the fundamental solution.

Real-World Example +

An e-commerce company's customer service director built a customer service assistant using Claude Projects. Her Instructions (System Prompt): "You are [Company Name]'s customer service assistant, specializing in order inquiries, return/exchange requests, and product questions. Language: Traditional Chinese. Tone: friendly and patient but not overly formal. Response length: concise; core information in the first three sentences. For out-of-scope questions (technical development, legal consultation), inform users it's outside your service scope and suggest other departments. Standard refund process: [company's refund policy]."

With this System Prompt, customer service staff just input the customer's question and Claude responds with the correct tone, format, and policy context — no need to re-explain the refund policy and tone requirements each time.

When to modify the prompt (not Instructions): for special customer situations, add supplemental context in the prompt — "This customer has bought three times and is always unsatisfied; they're a VIP; handle with extra care. Here's their question: [question]." This is for this special situation; no need to change Instructions.

When to modify Instructions: if the company's refund policy updates or tone requirements change, directly modify Instructions — all subsequent conversations automatically use the new settings.

Common Misconceptions +
✕ Misconception 1
× Misconception 1: System Prompt is a 'secret instruction' users can't see; Claude won't acknowledge its existence. System Prompt isn't secret — if you directly ask Claude "do you have a System Prompt?", it usually acknowledges yes (though it may not reveal specific content if the System Prompt requires confidentiality). System Prompt's 'invisibility' is because it doesn't display in the conversation interface, but Claude won't lie to protect a System Prompt. This aligns with AI honesty principles.
✕ Misconception 2
× Misconception 2: The more detailed the System Prompt, the better; write in every possible situation. Overly long System Prompts have two problems: higher costs (transmitted every call, directly affecting token costs); potentially worse performance (Claude may not effectively track all rules in very long System Prompts, especially conflicting ones). Most effective System Prompts: "clearly specify the most important rules, omit unimportant details" — this lets Claude clearly understand and follow your core requirements while maintaining flexibility for special situations.
The Missing Link +
Direct Impact

Core trade-off between System Prompt and in-conversation prompts: consistency vs flexibility. More rules in System Prompt makes conversation behavior more consistent and predictable (good for applications needing standardized output); but more rules reduce Claude's flexibility with special situations (may also block legitimately reasonable special requests). Best practice: put core, stable rules in System Prompt (role, language, basic format); dynamic, context-specific instructions pass in-conversation. This finds the best balance between consistency and flexibility.

Ask a Question
Please enter at least 10 characters