Additionally, non-clustered indexes on a clustered table use the clustered key as a pointer to locate data (instead of a direct physical pointer), which adds a bit of size to those indexes.

Why don't we have a pointer? Is it because the pointer can change when adding/removing rows and we don't want to update the non-clustered index?

However, storing the clustered key means when we search it, it will not be O(1), right? O(1) could be provided by a having pointer (talking only about the complexity of finding the row in the table once we have found the index value in the non-clustered index table).

Please explain this a bit more, it would be great having an example or even images for this specific non-clustered index on a clustered table case.