Prompt engineering / usage-help

we have two prompts:
A classification prompt that assigns legal practice areas to attorney profiles. A data research prompt used to extract structured information. When we test these prompts manually using the ChatGPT web interface, they consistently produce high-quality results.

When we run the same prompts through the API (using GPT-4o), the output is often: Incomplete
OR Inconsistent.

We’re using the same prompt and system message.

Guide us for this issue.

Try using this model: chatgpt-4o

ChatGPT-4o points to the GPT-4o snapshot currently used in ChatGPT.