Tell me about it... I work on prompting a lot and you would be surprised how different the model interprets things like "suggestion" vs "recommendation".