Adding to myself.
The Gemini programme was a mix of upgraded Mercury and pathfinder for Apollo. So it followed suit. For Apollo there is a need to be able to land almost anywhere around the planet. When coming from the moon you are committed to your landing site to a large extent at the moment you perform the burn to exit lunar orbit, some ability to fine tune the location is provided by correction burns on the way, but these take fuel and won’t give to the entire 24 hour rotation of the Earth to pick from. The Earth will be cheerfully rotating beneath you as you come home, and the point at the edge of the atmosphere where you must insert into re-entry is very very fine. The bit of geography underneath the entry point is determined by where the Earth has rotated to when you hit it, and nothing else. In the event you have an emergency and have to come home right away, there may be no option but to land half a world away from the usual site. The was a serious consideration during Apollo 13, although eventually they had the flexibility to have a reasonably easy landing. During Mercury they had no idea how bad things might become, and the option to bring the capsule down at a moments notice also played into planning.
Cecil’s original answer skates past a critical point of re-entry. He writes that the capsule slows with a retro-rocket, and then (if you were Russian) uses retro-rockets to cushion the landing. This ignores the fact that the actual speed is removed by aerodynamic forces in the atmosphere. To deorbit, the retro-rockets only re-orient the orbit enough that the capsule starts to dip into the atmosphere. On the return from the moon, even this isn’t done. The de-orbit burn at the moon points the capsule back to the Earth, and it slams into the atmosphere at full speed with only feeble attitude control rockets to perform manoeuvring. The Apollo capsule was a lifting body and had some modicum of control in the atmosphere to fly to a landing point, but not much. Enough that it could fly close to the pickup craft if the reentry point was good, but not enough to compensate for any change to re-entry point.
In the end the US has of course performed the vast majority of its manned landings onto the ground, and not the sea.