The same AI model, with two distinct prompts, can yield output of drastically different quality – one terrible and generic, the other precise and actually