Would it matter if the same prompt gives different output? You couldn't verify it.
How likely are you to just trust, let alone know for sure, whether or not the text I showed you is actually what I fed to the llm?