So, it’s important to get this right. In reputational effects, like this one, we don’t care about P(I use this | I’m a bad person) because, like you point out

So, it’s important to get this right. In reputational effects, like this one, we don’t care about P(I use this | I’m a bad person) because, like you point out, this obscures cases, like breathing air, where P(I use this) is already high. Instead, we care about P(I’m a bad person | I use this) And we especially care if this conditional probability is perceived to be high. This is because then your potential users will worry that if they use your product, others will make the (justified!) Bayesian inference that they are bad people. Because they don’t want to be seen to be bad people, they will avoid it.