“How likely” you would recommend something has always struck me as the wrong question to ask on a scaled question, especially when 0 means “Not at all likely.” Because anything above zero means there is a possibility of a recommendation.
In my research experience and my own personal experience the middle value on the scale is usually interpreted as neutral. Questions like “How likely” tend to be very confusing versus a version where 0 is something like “Very likely to advise against,” which fits better with the notion of retraction.
I agree NPS is nonsense. A zero could mean either all middle of the road responses, or extremely mixed results. A positive number above 50 would imply response skews positive and one below -50 skews negative. Everything in between is uninterpretable without analyzing the data.
I find that stakeholders understand a ratio of positive:neutral:negative responses much better. It gives a clearer view as to whether responses are in line or mixed and gives clearer direction for studying whether a feature or function is good or bad generally or if there are populations identified by a mixed response that may need to be addressed separately.