The same question gives a different list
SparkToro's testing found a very low chance of the same list of brands appearing twice when the same question is asked repeatedly. The ranking you see in one answer is one draw from many possible answers.
That makes a single check close to meaningless, and a score computed from a handful of checks looks more precise than it is.
How many answers is enough
Research on measuring AI answers argues for roughly seven to eight runs per prompt before a figure settles. This is why Slynnk gives every new prompt a ten-answer baseline on day one and then keeps sampling daily.
Whatever the number of answers, the honest way to show a result is the percentage with its likely range and the count beside it.
What to do with it
Read a visibility figure as a range, not a point. If two ranges overlap heavily, there is no clear difference.
Call a change real only when it is large enough and clearly bigger than that wobble. Slynnk requires at least 5 points and a statistical test, and says 'within noise' otherwise, so you do not chase a move that is not there.