I find it amusing that they tried to automate it with variables, etc, but seemingly wanted to make LLM generate a new response every time. Standard once written or even generated rejection form could've been more economically sound. But, but - then you wouldn't up the metrics of AI usage in your department!

Also email templates are a solved problem, there is like no reason to use AI for this

"Please generate an AI prompt for a polite rejection message for a candidate that wasn't picked"

I think the prompt might have been set up for like 1000 people, though.