First of all consider an AC coupled CRO, it will typically have a 1 Mohm input resistance with a .1 MF capacitor in series. If you apply a low frequency square wave to the input of the capacitor, the waveform across the resistor will not have a "flat" top, it will have have a slope. This slope will will go negative after a positive excursion and positive after a negative going edge. As the frequency of the squarewave gets lower, the slope gets worse and finally ends up as a spike where the signal returns to zero before the other edge come along.
The 1M ohm/.1MF input applies to the trigger input, so if the CRO was triggering on the slope bit of the wave form, any very slight variation of the actual triggering level would slide the trigger point point along the slope by a long way and the timebase would then then try and sychronise to a different part of the waveform and the waveform as view would be seen as multiple images sliding along the horizontal axis. if you then switched the triggering to DC, the top part of the squarewave would be flat and the trigger circuit would not be able to trigger from it at all, so you would naturally adjust the trigger level to halfway up the edge of the square wave and the CRO trace would lock up perfectly.
it is best to use AC coupling all the time except for low frequency waveforms. If you use DC coupling, the DC bias on you signal may fire the signal right of the screen, i.e. if you are trying to look at a 100mV signal sitting on a 12V DC level, you would set your sensitivity to 50 mV/CM to get a 2 CM high display but the DC level would deflect the display by 240 CMs and the Y shift will not be able to bring the trace back to the centre of the screen.
Frank