Probably these services are so tuned (not as in "fine-tuned" ML style) to each individual user that it's hard to get any sort of collective sense of what works and what doesn't. Not having any transparency what so ever into how they tune the model for individual users doesn't help either.
GPT4 and variants would only respond in vagaries, and had to be endlessly prompted forward,
Claude was the opposite, wrote actual code, answered in detail, zero vagueness, could appropriately re-write and hoist bits of code.