Ai node issue

Questions
1

Why is my AI node giving inconsistent response?

What is the error message (if any)?

Please share your workflow

(Select the nodes on your canvas and use the keyboard shortcuts CMD+C/CTRL+C and CMD+V/CTRL+V to copy and paste the workflow.)

Share the output returned by the last node

Information on your n8n setup

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
2

@Osas_Eduwoh LLMs are non-deterministic by default so some variation is normal, the first lever is temperature, drop it to 0 in the model nodes options for the most consistent output. but it depends what kind of inconsistent youre seeing, is the actual wording/content changing run to run, or is it the format/structure (the json shape, missing fields) thats varying? the fix is different for each, so which one is it?

3

High temperature settings, unclear prompts, or model limitations can cause variations. Try using more specific prompts and lower temperature values.

4

@Osas_Eduwoh

To add on, your prompt may also play a part.
Single-shot, multi-shot prompting may improve the output consistency

5

Hey! Inconsistent AI outputs usually come down to the model settings rather than an n8n bug. Because LLMs are probabilistic, they naturally drift unless you lock them down.

Try these quick fixes to get consistent results:

  1. Drop the Temperature: Lower it to 0 or 0.2 in your model node. This stops the AI from getting “creative” and forces strict responses.
  2. Hardcode the Output Format: In your system prompt, explicitly tell it how to answer. For example: “Output strictly in JSON. Do not include conversational filler like ‘Here is your data’.”
  3. Give Examples: Add 2 or 3 examples of a perfect input and output directly into the prompt (few-shot prompting). This gives the AI a clear pattern to copy.

If you can copy/paste your workflow here (just Ctrl+C on your canvas and Ctrl+V here) and share a quick example of a good vs. bad response, I can take a look and help you tweak it.