Do nothing when any action risks or costs more than leaving the page alone, and there are four clear cases where that holds. The first is a page already performing well. When a page ranks and converts, a change introduces risk for a gain that may not exist, so the disciplined move is to leave a winner alone rather than tinker. The second is rankings that are still settling or genuinely volatile. Acting on a position that is bouncing around means optimizing against noise, and you may “fix” something that was about to stabilize on its own.
The third case is a drop that traces to an in-progress update. When a broad change is rolling out and positions are moving across the board, the right response is often to wait it out rather than chase the moving target. Reacting mid-update can mean rewriting for a state that no longer exists by the time the rollout settles, and it makes it impossible to tell whether your change or the update caused the recovery. The fourth case is when the realistic return would not cover the effort. A page or term whose ceiling is modest does not justify a major rework, and the hours are better spent where the payoff is larger.
This runs against the activity bias that says you should always be optimizing. Motion is not the same as progress. Constant tinkering on stable or settling pages adds risk, muddies your read on what is actually working, and consumes time that could go to genuine opportunities. Choosing not to act is a real decision with its own logic, not laziness or neglect. The hard part is resisting the urge to do something simply because doing something feels productive, when holding is what the situation calls for.
Before you change anything, ask whether action risks or costs more than inaction. If the page is performing, if rankings are still settling, if an update is in progress, or if the return would not justify the effort, hold. Watch, let the situation resolve, and keep your attention on the places where action has a real and worthwhile payoff. Acting only when it earns its risk is the discipline here.