You select core utilization based on your experience with how well your designs route using a particular technology.
I think the default core utilization that FE uses is 70%. In my experience this value has worked pretty well across all the technologies that I've used. However, you may decide on a higher core utilization if, in your experience, your designs generally have plenty of free area and/or little routing congestion after post-route optimization.
It really depends on the technology, library, number of routing layers, etc. It can also depend on the timing and how much timing margin you want. For example, if you want to fix hold timing with 100ps extra margin, you will need more room for buffers. I typically start a block at about 65%. Sometimes we find we need a lower util, sometimes we find we can squeeze it tighter. It can also depend on how much of your design is std cells and how much is RAM or other IP.