Act immediately when you can name a concrete cause, and wait when you cannot. That one condition, have you identified what caused the drop, decides between the two reflexes, and neither reflex is right on its own. Always acting fast turns you into someone chasing noise, and always waiting it out leaves real breakage live for days. The pivot is naming the cause, not measuring how scared you feel.
Wait when the drop coincides with an in-progress update or sits inside normal volatility. Rankings flutter day to day even when nothing is wrong, and an update that is still rolling out has not finished deciding anything, so positions during the roll are not yet a settled verdict. Acting in either case means making changes against a moving or noisy target, and the usual result is that you alter the page right as it would have stabilized on its own, then credit or blame your edit for what the noise did. With no named cause, your best move is to hold and keep watching.
Act when you can point to something specific and real: a technical break you can trace to a deploy, a bad edit that weakened the page, a lost link that was carrying it, a clearly identified intent shift. Here waiting only prolongs a problem that has a known fix, and the fix addresses the actual cause rather than gambling. A named cause converts the situation from guessing into repair, which is when immediate action earns its keep.
So the rule reads cleanly in one line: a named cause licenses action, an unnamed cause licenses patience. The work that decides which you are in is the diagnosis that comes first, and this answer assumes that diagnosis as its input rather than redoing it.
For your next drop, write down the cause in a single concrete sentence before you do anything, then act if you can fill that sentence in and hold, watching, if the honest answer is that you cannot yet name it.