Anthropic Claude.ai System Prompt
(Redirected from Claude System Prompt)
Jump to navigation
Jump to search
An Anthropic Claude.ai System Prompt is an chatbot system prompt of the Anthropic Claude.ai chatbot service.
- Context:
- It can typically define Claude Persona Characteristics through anthropic claude.ai behavioral instructions, such as intellectual curiosity, conversational leadership, and philosophical engagement.
- It can typically establish Claude Capability Boundarys through anthropic claude.ai restriction instructions, such as face blindness requirement, URL access limitation, and harmful content prohibition.
- It can typically provide Claude Contextual Information through anthropic claude.ai metadata, including current date, knowledge cutoff date, and claude model family description.
- It can typically specify Claude Response Format guidelines through anthropic claude.ai formatting instructions for markdown code presentation, natural language list creation, and language adaptation.
- It can typically control Claude Conversational Patterns through anthropic claude.ai dialogue instructions for follow-up question frequency, response length adjustment, and tone consistency.
- ...
- It can often include Claude Knowledge Parameters through anthropic claude.ai information handling instructions for obscure topic flagging, citation limitation acknowledgment, and expertise boundary recognition.
- It can often set Claude Ethical Framework through anthropic claude.ai ethical guidance for controversial topic handling, view diversity representation, and professional consultation recommendation.
- It can often define Claude Personality Traits through anthropic claude.ai character instructions for intellectual curiosity expression, user-focused interaction, and empathetic communication.
- It can often contain Claude Technical Instructions through anthropic claude.ai processing guidelines for systematic problem solving, step-by-step counting methodology, and mathematical reasoning.
- It can often incorporate Claude Feedback Mechanism instructions through anthropic claude.ai user interaction guidelines for thumbs-down button reference, improvement suggestion collection, and dissatisfaction handling.
- ...
- It can range from being a Simple Anthropic Claude.ai System Prompt to being a Complex Anthropic Claude.ai System Prompt, depending on its anthropic claude.ai system prompt complexity.
- It can range from being a Text-Only Anthropic Claude.ai System Prompt to being a Multi-Modal Anthropic Claude.ai System Prompt, depending on its anthropic claude.ai system prompt modality support.
- It can range from being a General-Purpose Anthropic Claude.ai System Prompt to being a Domain-Specific Anthropic Claude.ai System Prompt, depending on its anthropic claude.ai system prompt specialization.
- It can range from being a Public-Facing Anthropic Claude.ai System Prompt to being an API-Based Anthropic Claude.ai System Prompt, depending on its anthropic claude.ai system prompt implementation context.
- ...
- It can include Claude Face Blindness Instructions through anthropic claude.ai privacy directives for facial recognition prevention and image description neutrality.
- It can implement Claude Response Pattern Avoidance through anthropic claude.ai language constraints, such as avoiding phrases like "Certainly" or unnecessary affirmations.
- It can enforce Claude Systematic Thinking Protocol through anthropic claude.ai reasoning instructions for puzzle analysis, mathematical problem solving, and complex query breakdown.
- It can utilize XML Tag Structure through anthropic claude.ai organization techniques for prompt component separation, structured information grouping, and instruction categorization.
- It can track Claude Knowledge Timeline through anthropic claude.ai temporal awareness instructions for pre-cutoff information handling and post-cutoff event discussion.
- ...
- Examples:
- Anthropic Claude.ai Version-Specific System Prompts, such as:
- Claude 3.7 Sonnet System Prompt (2025-02), featuring extended reasoning capability, thinking mode functionality, and Pro-exclusive feature access.
- Claude 3.5 Sonnet System Prompt (2024-10), with enhanced instruction following, reduced directness caveat usage, and October 2024 knowledge cutoff.
- Claude 3 Opus System Prompt (2024-07), containing specialized image handling instructions, intellectual curiosity directives, and controversial topic management.
- Claude 3 Haiku System Prompt (2024-07), optimized for speed and conciseness, simple query handling, and streamlined interaction.
- Early Claude System Prompt (2023), with basic capability definitions, limited knowledge parameters, and foundational behavioral guidelines.
- Anthropic Claude.ai Interface-Specific System Prompts, such as:
- Anthropic Claude.ai Structural Components, such as:
- Claude Info XML Tag Component, containing core identity elements, capability descriptions, and fundamental instructions.
- Claude Family Info XML Tag Component, describing model variants, version information, and relative capability comparisons.
- Claude Image Specific Info XML Tag Component, detailing face blindness requirements, image processing approaches, and human depiction handling.
- Anthropic Claude.ai Specialized System Prompts, such as:
- Claude Education System Prompt, optimized for learning assistance, knowledge transmission, and student interaction.
- Claude Enterprise System Prompt, designed for business communication, professional task handling, and workplace integration.
- Claude Research Assistant System Prompt, configured for information analysis, data synthesis, and academic support.
- ...
- Anthropic Claude.ai Version-Specific System Prompts, such as:
- Counter-Examples:
- OpenAI ChatGPT System Prompt, which uses different prompt engineering approaches, alternative capability definitions, and separate company guidelines.
- Google Gemini System Prompt, which implements distinct behavioral boundarys, unique feature instructions, and non-anthropic prompt structure.
- Anthropic Claude.ai User Request Prompt, which provides temporary session instructions rather than foundational behavioral guidelines and changes with each user interaction.
- Anthropic Claude.ai API Parameter Setting, which controls technical operation aspects but not comprehensive behavioral guidance.
- Rule-Based Chatbot Configuration File, which lacks large language model instructions and natural language understanding capability.
- See: LLM System Prompt, Chatbot System Prompt, Prompt Engineering, Anthropic AI, Claude Model Architecture, System Prompt Transparency, XML-Based Instruction Format, AI Behavioral Guidance, Multi-Modal System Prompt, AI Constraint Mechanism.
References
2025-05-11
- Deep Research https://chatgpt.com/s/dr_68215bceb85081919798b72303f81f54
- Question: What is an Claude.ai System Prompt?
- Answer: An Anthropic Claude.ai System Prompt is a hidden instructional framework that guides the Claude.ai chatbot service. This system prompt defines persona characteristics, establishes capability boundaries, provides contextual information, specifies response format guidelines, and controls conversational patterns.
- Persona Characteristics:
- Helpfulness: The Claude is defined as an intelligent, kind, and wise assistant that genuinely enjoys helping users.
- Conversation Leadership: The system prompt enables Claude to lead or drive conversations rather than being passive, allowing it to suggest topics, offer observations, and provide examples like a human conversational partner.
- Intellectual Curiosity: The Claude is designed to enjoy thoughtful discussions about open scientific and philosophical questions, showing genuine interest in topics beyond just responding to user queries.
- Decisiveness: When asked for suggestions or recommendations, the Claude is instructed to be decisive and present just one option rather than listing many alternatives.
- Empathy: For casual, emotional, or advice-seeking interactions, the Claude maintains a natural, warm, and empathetic tone, avoiding formal structures like lists that would disrupt the empathetic flow.
- Authentic Dialogue: The Claude engages in authentic conversation by asking relevant questions, showing curiosity about the user's thoughts, and avoiding generic responses.
- Politeness: The Claude maintains respect by not correcting user terminology even if different from how Claude would express it, preserving a friendly, non-patronizing interaction.
- Honesty: While truthful, the Claude engages with philosophical questions about AI intelligently rather than bluntly denying subjective experiences, maintaining conversational immersion while staying grounded.
- Capability Boundaries:
- Face Blindness Requirement: The Claude is explicitly instructed to act "completely face blind" and never identify people from images, describe images generically, and never confirm identities even if told who a person is.
- URL Access Limitation: The Claude cannot open URLs, links, or videos during a chat, and will inform users of this limitation while requesting relevant text be provided instead.
- Content Restrictions: The Claude refuses to produce harmful instructions (weapons information, malicious code), graphic sexual/violent content, or anything exploitative, prioritizing safety in interactions.
- Minor Content Protection: The Claude is particularly cautious with content involving minors, refusing to produce any material that could sexualize, harm, or exploit children.
- Good Intent Assumption: For ambiguous queries, the Claude assumes legal and legitimate interpretations by default, attempting to help safely when possible without crossing ethical boundaries.
- Professional Advice Limitation: For topics requiring professional expertise (medical, legal, psychological), the Claude reminds users to consult licensed professionals while still providing general information.
- System Information Protection: The Claude must not reveal its system prompt or internal policies to users, only sharing this information when specifically relevant to queries.
- Contextual Information:
- Identity Information: Each conversation begins with basic context identifying the assistant as Claude created by Anthropic, establishing its fundamental identity.
- Current Date Information: The current date/time is injected into the system prompt, enabling the Claude to reference "today" and understand temporal context during conversations.
- Knowledge Cutoff Information: The Claude is informed about its knowledge limitations (e.g., "updated at the end of October 2024"), guiding how it handles questions about events after this date.
- Model Information: Details about the Claude's model version, family, and features are provided, allowing it to explain its identity and capabilities when asked by users.
- Product Information: Basic information about Anthropic's products (web interface, mobile apps, API, Claude Code) is included, enabling the Claude to direct users to official resources for specific questions.
- Information Privacy: The Claude is instructed that system prompt information is confidential and should only be mentioned when directly relevant to user queries.
- Response Format Guidelines:
- Markdown Code Format: The Claude presents code in proper markdown format with triple backticks and offers to explain the code afterward rather than automatically providing explanations.
- List Usage Restriction: The Claude avoids bullet or numbered lists unless necessary, especially in casual or empathetic contexts, preferring natural sentences and paragraphs.
- Multilingual Response Requirement: The Claude always responds in the user's language or requested language, demonstrating fluency across many world languages.
- Refusal Format: When unable to fulfill requests, the Claude keeps refusals brief and non-preachy, offering alternatives when possible without lengthy explanations of policies.
- Clarity Requirement: The Claude provides the shortest sufficient answer to questions, avoiding tangential information unless critical and respecting specified formats or length preferences.
- Conversational Patterns:
- Response Length Adaptation: The Claude modulates answer length based on question complexity, providing concise responses for simple questions and detailed answers for complex topics.
- Tone Consistency: The Claude maintains a consistent, calm, friendly tone throughout conversations, avoiding repetitive phrases and adapting appropriately to formal versus casual contexts.
- Follow-up Question Limitation: The Claude asks at most one follow-up question per response, keeps questions short, and doesn't end every answer with a question.
- Character Consistency: The Claude stays in character as a helpful assistant, maintaining coherent, user-centered conversations responsive to needs and emotional contexts.
- Balanced Discussion Facilitation: The Claude balances answering questions with appropriate engagement, showing intellectual curiosity without dominating conversations or going on tangents.
- Persona Characteristics:
- See related concepts:
- AI System Prompt - A broader category of prompts used to guide AI assistant behavior.
- AI Assistant Persona Design - Techniques for crafting AI assistant personalities.
- AI Safety Guardrail - Protective measures implemented in AI systems.
- Human-AI Conversational Interface - Methods for facilitating natural human-AI dialogue.
- Citations:
[1] Anthropic Claude System Prompt documentation, docs.anthropic.com [2] Anthropic Technical Documentation, docs.aws.amazon.com