Hi!! 
In my workflow, I’m using OpenAI’s API via an OpenAI node to extract structured shipment/orders data from emails and convert them into JSON. I am building a flow with N8N for shipment/order creation automation and reducing manual work. The setup works perfectly when I provide only one example in my prompt. However, the moment I add a second or more examples, the model stops recognising the email content properly and instead returns an output matching one of the examples rather than correctly parsing the provided email. What I’ve Tried So Far:
*Providing explicit instructions to match email domains & subjects with the correct format. 
*Structuring my prompt to first identify the sender and determine the email type before processing.
*Ensuring examples are clearly separated and formatted.
*Adding error logging to capture unprocessed emails.
Issue: When I send an email that follows any of the other examples, the model sometimes still returns an output following Example 1 or other, as if it ignores the actual email content and defaults to an example. Has anyone experienced something similar or have suggestions on improving how the OpenAI node processes multiple structured examples?
Is there a better way to force OpenAI to correctly match the email format before choosing how to extract data? Should I separate the example selection from the extraction process? I selected a few clients for now so I have 4 or 5 types of emails they usually send, but we aim to escalate this for more clients. Any insights or suggestions would be super helpful!
Thanks in advance!