First of all, I'm using KiCAD, but I guess the same problematic exists in any layout software.
When I need to create a custom footprint for an SMD component, I only set the coordinates of whatever this footprint consists of (pads, drawings, etc.) relative to the footprint editor's grid origin. When I generate the Pick'n'Place file for the final board, KiCAD uses those grid origins for the components' coordinates. That means, that the offset I might introduce when creating the footprint matters a lot. In fact, if I move all the shapes in a footprint relatively to the origin in the footprint editor, I will get different component positions in the final board while the layout is the same (namely, the gerber files are the same).
In my special case, I have the MOLEX-5025850570 with this recommended landing pattern:
In this case, it's pretty obvious that the x-origin is in the horizontal center, because the component is symmetrical. But I don't see how the y-origin is determined. Below are two versions of the footprint which will both result in a correct layout in the final board, but with different pick-place files.
The assembler has a similar problem (how does he know at which point the pick-place machine has to pick up a component from the tape), but for this, the datasheet specifies a "center of connector" in the packaging section.
But how do I know where this position is in the smd landing pattern? I would have expected this information to be there.
I'm not too familiar with the pick and place process in practice: do assemblers typically make a test run and adjust footprint positions manually or do they completely rely on the position files to be absolutely precise?