Accuracy is best thought as relative, though the usage of percentages here is not a good fit imho. You compare your result to the wanted value by dividing the error by the value. In this case |x - √2|/√2 < 0.000001 (or 0.0001% if you insist)

For comparison, the initial value of the iteration (1.2) would be 85% accurate, which without context might sounds like a lot but is pretty abysmal