Generally I agree, but in this case it's very easily found.
pm24601 to shitgoose: why!?
shitgoose to pm24601: go google it yourself!
pm24601 to shitgoose: why should i google why i suck?
shitgoose to pm24601: if you choose to silently disbelieve that is fine. Expecting anyone else to spend their time and energy on "proving" some thing to someone else, should only happen if there is a pay off.
Occam's razor doesn't really apply to determining an individual account, as there's no model difference in a particular account being either, given many accounts and each being some.
You should respond based on considering it a superposition of both (and more exotic states, like a non-shill blindly repeating shill points), based on the distribution of likelihood given their speech patterns, background prevalence, etc.
That's really the simplest model which lets you incorporate all the information you know in crafting a response -- not a false dichotomy of making assumptions you know will sometimes be erroneous. (...Which is not what Occam's Razor is about.)