n8n AI Agents Tutorial: System & User Prompts for Consistent AI Agents (2026)
Building an AI agent in n8n is straightforward. The challenge is writing effective prompts. While connecting the AI Agent Node to a chat model is quick, your prompts decide the agent’s performance. This guide covers System and User Prompts, with examples and a framework for n8n.
The n8n AI Agent Node
The n8n AI Agent Node is a workflow component. It links Large Language Models such as OpenAI GPT or Google Gemini to tools, memory, and structured prompts for task execution. You can create AI chatbots, data processors, or assistants on the n8n platform without coding. Key parts are the node, a chat model, optional tools for external actions, and Simple Memory for conversation history.
System vs. User Prompts
In n8n’s AI Agent setup, System Prompts give behavioral instructions. User Prompts hold specific requests. User Prompts can vary. The System Prompt stays constant, forming the agent’s personality and rules.
Example: Baseball Historian
System Prompt: "You are a baseball historian who researches based on questions provided. If you're unsure about a question, state you do not know. If it's an opinion-based question, state that as well since it is not a fact."
User Prompt: "What are three of the weirdest baseball fields used in the MLB during the early 1900s?"Components
System Prompt Components:
1. Role/Persona: “You are a baseball historian” – Sets identity and expertise.
2. Rules/Restrictions: “who researches based on questions provided” – Defines the working method.
3. Error Handling: “If you’re unsure about a question, state you do not know” – Manages knowledge gaps.
4. Contextualization: “If it’s an opinion-based question, state that as well since it is not a fact” – Separates facts from opinions.
User Prompt Components:
1. Task: “What are three of the weirdest baseball fields” – The specific request.
2. Context: “used in the MLB during the early 1900s” – Limits time and institution.
3. Output Format: Implicit: A list of three examples with descriptions.
4. Constraints: “weirdest” – A subjective criterion needing historical knowledge.
Implementation in n8n
User Prompt via Chat Trigger
Workflow Structure:
1. Chat Trigger Node → 2. AI Agent Node (Prompt: {{ $json.chatInput }}) → 3. Chat Model Connection
Configuration:
- Prompt Source: 'Chat Input'
- Dynamic Variable: {{ $json.chatInput }}
- Optional: Simple Memory for conversation historyThis setup works for interactive chatbots. The User Prompt comes from user input. The expression {{ $json.chatInput }} pulls text from the Chat Trigger.
Defined User Prompt with Data Integration
Workflow Structure:
1. Edit Fields Node (boxer: "Jack Johnson") → 2. AI Agent Node
AI Agent Configuration:
- Prompt Source: 'Define Below'
- Prompt Text: "Based on the provided boxer {{ $json.boxer }}, can you give me their win and loss record?"
- Confirm Expression: Press Enter after {{ $json.boxer }}This pattern helps when processing data from earlier workflow nodes. Dynamic data injection via {{ $json.fieldName }} enables automated pipelines.
Advanced System Prompt with Tools
System Prompt with Tool Integration:
"You are a marathon running coach that helps runners improve their times to hit Boston qualifying time.
TOOLS:
- Calculator: Use for pace calculations. Example: '40-minute 10K is how many minutes per mile?'
- Strava API: Use to fetch recent training data when asked about performance history.
ERROR HANDLING:
- If the Strava API is not working, state: 'Error with the Strava API.'
RULES:
- Only give recommendations based around proven data.
- If a question is outside of running, say you cannot answer it.
- If you need more information, ask clarifying questions.
TONE:
- Use a tone that you would find with a workout buddy or partner."This System Prompt uses advanced methods: clear tool descriptions with examples, proactive error handling, defined content limits, and communication style details.
AI Agent System Prompt Framework
A structured System Prompt can have up to seven parts:
1. ROLE/IDENTITY: "You are a [specific role] that [main function]."
2. TOOLS: "You have access to [tool list]. Use [tool] when [use case]."
3. TONE: "Communicate in a [tone] manner, similar to [comparison]."
4. RULES/RESTRICTIONS: "Always [behavior], never [prohibition], if [condition] then [action]."
5. EXAMPLES: "Q: [example question] A: [example answer]" (Few-Shot Prompting)
6. OUTPUT FORMAT: "Respond in [format] with [structure]. Include [elements]."
7. ERROR HANDLING: "If [error state], then [error handling]."Be specific. Detailed components lead to more consistent agent responses. Avoid generic descriptions like “helpful assistant”. Use concrete, task-focused identities.
Frequently Asked Questions
Can a user override the System Prompt in n8n?
No. The System Prompt is fixed in the AI Agent Node’s configuration. User inputs cannot alter it. This maintains consistent behavior.
How many tools should an AI Agent have at most?
2-3 well-described tools work best. An agent with 15+ tools often gets confused and selects incorrectly. Each tool’s description in the System Prompt should include its exact name, primary use, and a concrete example. The prompt name must match the n8n configuration.
What are common mistakes with System Prompts?
1. Too vague role definition: “Helpful assistant” instead of “Specialized financial analyst with focus on retirement planning”.
2. Missing error handling: No instructions for API errors or data gaps.
3. Contradictory instructions: “Always be concise” but “Provide comprehensive details”.
4. Overloaded tool lists: Too many options without clear prioritization.
5. Neglected output formats: No specification for structured responses.
How do I implement Few-Shot Prompting in n8n?
System Prompt with Few-Shot Examples:
"You are a running coach. Here are examples of good responses:
Q: Is it okay to add speedwork to a long run?
A: Absolutely. This is a common approach in Fitz and Daniels training plans...
Q: How do I calculate my lactate threshold?
A: You can estimate it by...
Now answer the user's question following these examples."Few-Shot Prompting with 2-3 examples reduces hallucinations and sets consistent response patterns. Examples should cover typical use cases and show different answer types.
Can I reuse predefined System Prompts from n8n AI Nodes?
Yes. Several specialized n8n AI Nodes contain good System Prompt templates. The Sentiment Analysis Node categorizes text. The Text Classifier offers classification prompts. The Guardrails Node has detailed format and safety rules. These can be a starting point for adaptation. The Guardrails Node shows particularly professional prompt structures.
How do I handle dynamic context data in the System Prompt?
The System Prompt itself is static. Context can be added via the User Prompt or workflow data. For truly dynamic aspects, like daily role changes, workflow logic is required to recreate the entire AI Agent Node with a new System Prompt. Usually, a static System Prompt with a dynamic User Prompt is the practical choice.
Source
Based on this article.
