Why is the O1 system prompt ineffective in non-conversational nodes in n8n?

Below is an explanation tailored around n8n's current behavior and documentation:

Why the System Prompt Works Only in Conversational Setup

• Designed for Chat Models:
  In n8n’s LangChain-based nodes (such as the Basic LLM Chain node), the system prompt (or system message) is designed to be used as part of a structured conversational input. This input typically includes separate fields for system, user, and AI messages.

• Non-Conversational Nodes Ignore It:
  When you aren’t using a chat model (i.e., not connected to a node that expects chat messages), the node won’t process your system prompt. In those cases, only a plain text prompt is considered. That’s why the system prompt isn’t applied outside the intended, conversational context.

• Output Quality Impact:
  If you try to force the system prompt in a non-conversational setup, the node might generate output that does not reflect the contextual cues intended by a structured conversation. This may be why you’re experiencing suboptimal or “bad” outputs.

Recommendation

• Set Up as a Conversational Node:
  Ensure you are using a chat-capable node and that you configure the system prompt along with user and AI messages. This structure helps guide the model by giving clear instructions. For example, a system message might instruct the model on tone or perspective (e.g., “Always respond like a pirate”), while user messages provide the query.

• Check Your Configuration:

  • Make sure you are not inadvertently mixing non-conversational and conversational settings.
  • Verify that your node is connected to a compatible chat model if you wish to leverage system messages.

By aligning your configuration with the intended usage of chat nodes in n8n, you should see an improvement in how context is handled, resulting in better output.