The ChatGPT AIs are provided a bit of information you don’t see in a system role. Their system role message tells them information similar to “You are ChatGPT, based on GPT-4 from OpenAI…”
(that information about the model is actually not that reliable - if you press ChatGPT for details about how it knows what model it is, it will fall back to its pre-training which of course doesn’t have any info about GPT-4 being released)
The API requires you to construct your own system role message to direct the operation of the AI. You can tell it “You are ChatMonkey, an AI based on a collection of monkey neurons in a jar in a mad scientist’s lab”, and it will try to fulfill user questions with the information at hand.