number of elements in the path from new leaf to root $= \log n$, and all elements are sorted from leaf to root so we can do a binary search which will result in $O(\log \log n)$ number of comparisons.
Since in heap is a complete binary tree, in an array implementation, from every node index, we can know its depth and this will be the $n$ for binary search.