I find it telling that I have (mostly) good experiences with the GPT family and (mostly) bad experiences with the Claude family. I just wish I could figure out

I find it telling that I have (mostly) good experiences with the GPT family and (mostly) bad experiences with the Claude family. I just wish I could figure out what it tells. Their training data can't be that different. The problems I'm feeding them are the same. Many people think Claude is the more capable of the two. It has to be how I'm presenting the problems, right? What other variable is there?