I did a bit of research and even water just "involved" in the process is counted as used in this context. For instance river water that is used for cooling and returned is counted.
I think these sort of graphs are simply misleading and should not be used.
Using natural water is more common in power plants, where the volume of water is sufficiently large to use concrete aquaducts and the precision of the temperature isn't as important. Most datacenter liquid cooling is treated municipal water with glycol additives to prevent corrosion and pump jamming.
Is that actually a common configuration? I could find very little on data centers being cooler by river water, and the few I found sounded like they were doing something novel.