Personally, I would not use a 3-point scale if you intend to publish the research in the future. Some journals are now even requiring motivation for the use of a 5-point scale (instead of a 7-point scale).
However, a 5-point scale is self-explanatory and commonly used, so I do not believe that respondents will have difficulty in distinguishing between the 5-points.
There is hardly any right or wrong in these matters. Take a look at similar scales in whatever field you work in and get an idea of what kind of response format the use. However, in order to reasonably treat a Likert-scale as metric, four to five response options ought to be provided.
It depends upon the quality of respondents, as while developing achievement test questions, Linn (2008) recommends to develop direct questions for primary graders because they will not understand the statements or hypothesis by items. Similarly, it the respondents are better enough to think more critically then use the 5 or 7 point likert scale to provide them opportunity to respond as per their own level. And if your respondents are children or uneducated people then use the 3 point scale because your should not put them in confusion.