System prompt
Also written: Monolithic system prompts, System instructions, Meta prompt
Core instructions given to an AI to define its behavior, persona, and rules for all interactions.
A system prompt is like the AI's rulebook. It's a set of instructions given to the AI model before it starts interacting with you. These instructions tell the AI how to behave, what kind of personality to adopt, and what rules it must follow.
Think of it like giving a character in a play their script and stage directions. The system prompt ensures the AI stays on track and acts consistently. For example, a system prompt might tell an AI to always respond politely, never reveal its internal workings, or to act as a helpful assistant for a specific task.
The system prompt is a foundational set of instructions provided to a large language model (LLM) to define its operational parameters, persona, and constraints. It guides the model's behavior across an entire session, setting the context for all subsequent user inputs and model outputs. This prompt is distinct from user prompts, which are specific queries or commands.
Effective system prompts are crucial for controlling LLM behavior, ensuring safety, and aligning the model with desired tasks or personas. Developers use them to establish guardrails, define the AI's role (e.g., chatbot, summarizer, code generator), and prevent undesirable outputs. The quality and specificity of the system prompt significantly impact the model's performance and reliability.
System prompts are crucial for defining an AI's behavior and ensuring it follows specific rules or adopts a particular persona. Recent discussions highlight their importance in training models for security, like OpenAI's work on prioritizing instructions, and in building more reliable AI systems, as seen with Google's ADK 2.0 workflows.
- Google ADK 2.0 workflows tutorial: Building reliable multi-agent systemsGoogle · Google Cloud Tech on YouTube ·
- Building scalable AI agents with modular prompt transpilationGoogle · Google Developers Blog ·
- FIXING Opus 5: PROOF that Prompt Engineering IS NOT DEADAnthropic · IndyDevDan ·
- Grok Bot is not what you thinkxAI · Ben's Bites ·
- Opus 5 >> Fable 5OpenAI · Ben's Bites ·
- Project Vend: Can Claude run a small shop? (And why does that matter?)Anthropic · Anthropic Research ·
- How we are building the personal health coachGoogle · Google Research Blog ·
- The Instruction Hierarchy: Training LLMs to Prioritize Privileged InstructionsOpenAI · OpenAI News ·
- Anthropic Claude
- Anthropic's work on Opus 5 suggests that prompt engineering, which includes system prompts, remains vital for achieving better AI performance.
- Google Gemini
- Google's ADK 2.0 workflows aim to move beyond complex prompts, implying a structured approach to guiding AI behavior, likely involving system prompts.
- OpenAI ChatGPT
- OpenAI has focused on training ChatGPT to prioritize system instructions, enhancing security against malicious inputs and ensuring reliable behavior.
- User prompt
- A user prompt is a specific question or command given by the user t
- Fine-tuning
- Fine-tuning adjusts the model's weights based on specific data, whi
- Few-shot learning
- Few-shot learning involves providing examples within the prom
- Reinforcement learning from human feedback (RLHF)
- RLHF is a training technique
The concept of system prompts emerged with the development of large language models, formalized as a way to control model behavior beyond simple user inputs. Early research and practical applications in chatbot development established its utility.
Open ChatGPT, Claude, or Grok and start a new chat. Before asking anything else, type a message like: 'Act as a pirate captain. Respond to all my questions as if you were a pirate from the 17th century. Use pirate slang.' Then, ask it a simple question like 'What is the weather today?' Observe how it adopts the persona.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Anti-patternAvoid monolithic system prompts for AI agentsMonolithic system prompts lead to scaling bottlenecks and runtime errors; modularize instead. · 10 min
- Prompt patternUse positive and negative patterns in system promptsDefine what to replicate and what to avoid to shape AI output precisely. · 15 min
- System promptClaude's system prompt for managing an automated shopInstructions for Claude to manage inventory, pricing, and customer interactions for a small automated store. · 10 min
- System promptPersonify Grok agents with system prompts for specific rolesDefine agent roles like 'money manager' to guide their behavior and interactions. · 5 min
- Troubleshooting fixFix verbose Opus 5 with system prompt engineeringUse system prompts to make Opus 5 precise and reduce token waste. · 15 min
Questions people ask
- Can I change the system prompt after the AI has started?
- Generally, you cannot change the system prompt mid-conversation without starting a new chat. The system prompt sets the initial behavior for the entire session.
- Is the system prompt the same as the first message I send?
- No, the system prompt is usually set by the developers or the platform behind the AI and is invisible to you. Your first message is typically a user prompt.
- Why are system prompts important?
- They are crucial for controlling the AI's behavior, ensuring it stays within desired boundaries, adopts a specific persona, and performs tasks reliably and safely.
First seen on Subvolts Apr 19, 2024. Updated Oct 5, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.