How you frame a question changes what an LLM actually argues, not just its tone

I ran a two-minute test on three different AI assistants last week that I can't stop thinking...