I don't think that there are any official guidelines out there which give definite acceptance criteria for accuracy. And this is actually good, because acceptance criteria for accuracy (and any other validation parameter) can be wildly different depending on the purpose of the respective analytical method.
Regarding the +-2% you mentioned, this is quite a common criterion for validation of assay tests in the pharmaceutical field. But for related substances, things are VERY different! Just think of it: at the 0.1% level defining an acceptable range of +-2% would mean you'd have to hit the 0.098-0.102% range with your recovery - impossible! I'd say there's no pharmaceutical related substance method out there which is precise and accurate enough to distinguish between these concentrations. An actually it doesn't have to.
I've seen a lot of validations of related substances methods, and usual acceptance criteria for accuracy via recovery are 90-110% or even 80-120%, especially if you look at lower concentrations close to the LOQ.