My solution to this kind of problem is not to use a tag in this way at all. It’s a complete misuse of tags (in my opinion), and is likely to cause trouble down the road (as you’ve discovered), because cells are reused.
Typically, the problem being solved is this: A piece of interface in a cell is interacted with by the user (e.g. a button is tapped), and now we want to know what row that cell currently corresponds to so that we can respond with respect to the corresponding data model.
The way I solve this in my apps is, when the button is tapped or whatever and I receive a control event or delegate event from it, to walk up the view hierarchy from that piece of the interface (the button or whatever) until I come to the cell, and then call the table view’s indexPath(for:), which takes a cell and returns the corresponding index path. The control event or delegate event always includes the interface object as a parameter, so it is easy to get from that to the cell and from there to the row.
Thus, for example:
UIView* v = // sender, the interface object
do {
v = v.superview;
} while (![v isKindOfClass: [UITableViewCell class]]);
UITableViewCell* cell = (UITableViewCell*)v;
NSIndexPath* ip = [self.tableView indexPathForCell:cell];
// and now we know the row (ip.row)
[NOTE A possible alternative would be to use a custom cell subclass in which you have a special property where you store the row in cellForRowAt. But this seems to me completely unnecessary, seeing as indexPath(for:) gives you exactly that same information! On the other hand, there is no indexPath(for:) for a header/footer, so in that case I do use a custom subclass that stores the section number, as in this example (see the implementation of viewForHeaderInSection).]