Okay here's how I made it work realiably:
-Set the image points of elements to their top left corners
-Create the HTML element at 0,0 coordinates
-Take the Construct editor coordinate of the target object (to which you want to attach an HTML element)
-Divide the x coordinate with 16 and the y coordinate with 16 in order to get em's
It would be really nice to have some built-in way of "anchoring" HTML elements to regular object instances!