According to the Apollo 14 LM Timeline Book, the MODE SEL switch that selects the source for the pointers is set at two different points.
At LANDING RADAR CHECKOUT, at mission time 106:30, the RATE ERR MON switch is set to LDG RDR/CMPTR and MODE SEL is set to LR, so the cross-pointers are controlled from the landing radar.
Between 108:00 and 108:20, we have UPDATE FROM MSFN -- a data uplink from the ground, presumably updating the spacecraft's computer with the latest and greatest position and velocity information. During this sequence, MODE SEL goes to PGNS; the pointers are controlled by the primary guidance computer.
Powered descent starts at 108:40 with the MODE SEL still in PGNS mode. I don't see any other mentions of MODE SEL in the timeline, but searching OCR'd PDFs can be error prone, so I can't rule out some other mode change in there, but I think nominal landings were done with the cross-pointers driven by PGNS.
In Digital Apollo, I noticed something rather interesting in the discussion of the 1201/1202 errors encountered during the Apollo 11 landing. The notes for those errors in the program alarm "cheat sheet" say:
Watch for other cues, PGNCS condition unknown, DSKY may be locked up, duty cycle may be up to point of missing some functions (nav. last to die) Switch to AGS (follow ERR needles) may help (reduces PGNCS duty cycle signif.)
In other words, when hitting the executive overflow errors that Apollo 11 encountered, it was suggested to switch the cross pointer needles to AGS-driven to lighten the load on the primary computer. No such suggestion was relayed to Armstrong and Aldrin at the time, however; Aldrin noticed the errors seemed to occur when a particular data display was up, so they simply monitored that datum from the ground instead for the remainder of the descent.