1) Number of cells present at the time of 100% confluency..
2) Dividing time for the cells.
3) Treatment, whether it related to cytotoxicity/cell inhibition or cell proliferation.
As a thumb rule , u should not have 80% of the confluency in the end of the experiment. Therefore, depending upon the the these four factors, you can decide how many cells you have to seed.
If you are seeding more no. of cells then cells will start growing on each other when they don't have the space to grow. Due to which the cell present on the lower surface don't get the the proper treatment and you will get the false positive results.
Of not, make sure that when you seed the cells they should be in single cell suspension otherwise they will grow in patches/clumps.
It depends on many factors actually. But in general, you should know the seeding density of cells from previous starting assays such as cell growth and viability assays. For instance, if you seeded 5,000 or 10,000 cells/well in a 96 well plate and prove that your drug is cytotoxic on cells, while control group remains healthy up to three days, then you can seed 500,000 or 1 million cells per 100 mm culture dish, respectively- knowing that by three days, your control group will remain healthy while your treatment group is affected. It is a matter of area ratio, and number of cells being seeded in early experiments. As long as your cells are happy and not forming clumps, while still within the acceptable range of cell confluence (maximum 80%), you should be fine.