In VASP it is hardly an added effort, around 10-20% more time, and I might be exaggerating.

anyway, usually when you do GGA+U calculation,  do you insert +U from the beginning (atomic relaxation) or do you use GGA only for the atomic relaxation insert U when you calculate the band gap?

You can do either, but if GGA+U is a good descriptor of your compound than its pretty safe to use it all the way, usually it gives very good geometries in agreement with experiment, provided you use the correct U value. In some cases the difference in geometry might be insignificant, although there is always an option this might be seriously different. You should try several options and see what is most appropriate, but usually it is safer to find the electronic structure in a state where the structure is fully relaxed, and you only give up on such luxuries with extremely expensive theories, like hybrid DFT or GW.